notebooks_and_scripts/composite_spectrum_tests/match_tests.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "90243e0e-c767-466f-ba0d-1e540f3a036c",
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib import pyplot as plt, colors as mcolors, colormaps\n",
    "import numpy as np\n",
    "from scipy.spatial.distance import cosine as cos_dist\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "69a3985d-1b66-4410-b68f-cb6fe48a3d1c",
   "metadata": {},
   "outputs": [],
   "source": [
    "color = lambda cmap, vmin, vmax, v: colormaps[cmap](mcolors.Normalize(vmin=vmin, vmax=vmax)(v))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "35515ed1-aa14-4654-a354-c309cab7ee02",
   "metadata": {},
   "outputs": [],
   "source": [
    "def spec_norm(spectrum):\n",
    "    \"\"\" normalize spectrum such that all intensities sum to 1 \"\"\"\n",
    "    mz, i = np.array(spectrum)\n",
    "    return np.array([mz, i / sum(i)])\n",
    "\n",
    "\n",
    "def spec_entropy(spectrum):\n",
    "    \"\"\" compute spectral entropy for single spectrum \"\"\"\n",
    "    mz, i = spectrum\n",
    "    return -np.sum(i * np.log(i + 0.001))\n",
    "    \n",
    "\n",
    "def spec_combine(spectra, comb_mztol=0.1):\n",
    "    \"\"\" combine multiple spectra into a single spectrum \"\"\"\n",
    "    comb_mz, comb_i = [], []\n",
    "    # iterate through spectra in set\n",
    "    # combine m/zs that are within tolerance of one another\n",
    "    # otherwise just add the new points\n",
    "    for spectrum in spectra:\n",
    "        for mz, i in zip(*spectrum):\n",
    "            added = False\n",
    "            for j in range(len(comb_mz)):\n",
    "                if abs(mz - comb_mz[j]) <= comb_mztol:\n",
    "                    comb_i[j] += i\n",
    "                    added = True\n",
    "            if not added:\n",
    "                comb_mz.append(mz)\n",
    "                comb_i.append(i)\n",
    "    # sort by m/z\n",
    "    idx = np.argsort(comb_mz)\n",
    "    comb_mz, comb_i = np.array(comb_mz)[idx], np.array(comb_i)[idx]\n",
    "    # return normalized spectrum\n",
    "    return spec_norm([comb_mz, comb_i])\n",
    "\n",
    "\n",
    "def spec_entropy_similarity(spectrum_A, spectrum_B):\n",
    "    \"\"\" pairwise spectral entropy based similarity as defined in the paper \"\"\"\n",
    "    spectrum_AB = spec_combine([spectrum_A, spectrum_B])\n",
    "    s_AB, s_A, s_B = spec_entropy(spectrum_AB), spec_entropy(spectrum_A), spec_entropy(spectrum_B)\n",
    "    return 1. - ((2. * s_AB - s_A - s_B) / np.log(4.))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "e03eddbd-e3c6-46af-b039-73a538f57a5d",
   "metadata": {},
   "outputs": [],
   "source": [
    "import sqlite3\n",
    "\n",
    "con_nist = sqlite3.connect('nist20_subset.db')\n",
    "con_nist.row_factory = sqlite3.Row\n",
    "cur_nist = con_nist.cursor()\n",
    "\n",
    "con_mona = sqlite3.connect('MoNA_subset.db')\n",
    "con_mona.row_factory = sqlite3.Row\n",
    "cur_mona = con_mona.cursor()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "86f5d0ab",
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "threshold: 0.65\n",
      "\tsampled: 719\n",
      "\ttp: 4979 fp: 43087 tn: 522896 fn: 3287\n",
      "threshold: 0.55\n",
      "\tsampled: 736\n",
      "\ttp: 5624 fp: 65154 tn: 507520 fn: 2825\n",
      "threshold: 0.45\n",
      "\tsampled: 753\n",
      "\ttp: 6681 fp: 89719 tn: 480263 fn: 2246\n",
      "threshold: 0.4\n",
      "\tsampled: 725\n",
      "\ttp: 6564 fp: 99693 tn: 459792 fn: 2030\n",
      "threshold: 0.35\n",
      "\tsampled: 730\n",
      "\ttp: 6617 fp: 121403 tn: 433916 fn: 1687\n",
      "threshold: 0.3\n",
      "\tsampled: 780\n",
      "\ttp: 7499 fp: 152790 tn: 444229 fn: 1592\n",
      "threshold: 0.25\n",
      "\tsampled: 748\n",
      "\ttp: 7544 fp: 194040 tn: 395606 fn: 1163\n",
      "threshold: 0.2\n",
      "\tsampled: 779\n",
      "\ttp: 8224 fp: 226605 tn: 379886 fn: 1002\n",
      "threshold: 0.15\n",
      "\tsampled: 725\n",
      "\ttp: 7843 fp: 265887 tn: 303277 fn: 524\n",
      "threshold: 0.1\n",
      "\tsampled: 795\n",
      "\ttp: 8754 fp: 367366 tn: 254352 fn: 434\n",
      "threshold: 0.05\n",
      "\tsampled: 722\n",
      "\ttp: 8428 fp: 398768 tn: 150005 fn: 121\n"
     ]
    }
   ],
   "source": [
    "# randomly sample a subset of the database\n",
    "np.random.seed(425)\n",
    "\n",
    "# test classification accuracy using complete database\n",
    "metrics_rep13 = {}\n",
    "for threshold in [0.65, 0.55, 0.45, 0.4, 0.35, 0.3, 0.25, 0.2, 0.15, 0.1, 0.05]:\n",
    "    print('threshold:', threshold)\n",
    "    tp, fp, tn, fn = 0, 0, 0, 0\n",
    "    qry_sel1 = 'SELECT name, adduct, pre_mz, msms_mz, msms_i FROM msms;'\n",
    "    qry_sel2 = 'SELECT name, adduct, msms_mz, msms_i FROM msms WHERE pre_mz>=? AND pre_mz<=?;'\n",
    "    i = 0\n",
    "    for name, adduct, pre_mz, msms_mz, msms_i in cur_nist.execute(qry_sel1).fetchall():\n",
    "        if np.random.random() >= 1 / 500:  # sample around 1000 per threshold\n",
    "            # skip this iteration of the loop\n",
    "            continue\n",
    "        mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "        spectrum = spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                              np.fromstring(msms_i.strip('[]'), sep=',')])\n",
    "        for name2, adduct2, msms_mz2, msms_i2 in cur_nist.execute(qry_sel2, (mz_min, mz_max)).fetchall():\n",
    "            spectrum2 = spec_norm([np.fromstring(msms_mz2.strip('[]'), sep=','), \n",
    "                                   np.fromstring(msms_i2.strip('[]'), sep=',')])\n",
    "            # label based on similarity threshold\n",
    "            label = spec_entropy_similarity(spectrum, spectrum2) >= threshold\n",
    "            # actual ground truth\n",
    "            known = (name == name2) and (adduct == adduct2)\n",
    "            # count TP, FP, TN, or FN\n",
    "            if label:\n",
    "                if known:\n",
    "                    tp += 1\n",
    "                else:\n",
    "                    fp += 1\n",
    "            else:\n",
    "                if known:\n",
    "                    fn += 1\n",
    "                else:\n",
    "                    tn += 1\n",
    "        i += 1\n",
    "    print('\\tsampled:', i)\n",
    "    print('\\ttp:', tp, 'fp:', fp, 'tn:', tn, 'fn:', fn)\n",
    "    metrics_rep13[threshold] = {'sampled': i, 'tp': tp, 'fp': fp, 'tn': tn, 'fn': fn}\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "96b35410",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "with open('full_match_test_2.json', 'w') as jf:\n",
    "    json.dump([metrics_rep11, metrics_rep12, metrics_rep13], jf, indent=4)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "6d56a889",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "threshold: 0.65\n",
      "\tsampled: 736\n",
      "\ttp: 4359 fp: 45272 tn: 522360 fn: 3642\n",
      "threshold: 0.55\n",
      "\tsampled: 762\n",
      "\ttp: 5326 fp: 65452 tn: 518677 fn: 3010\n",
      "threshold: 0.45\n",
      "\tsampled: 739\n",
      "\ttp: 5691 fp: 88704 tn: 499068 fn: 2387\n",
      "threshold: 0.4\n",
      "\tsampled: 701\n",
      "\ttp: 5577 fp: 101970 tn: 453808 fn: 1891\n",
      "threshold: 0.35\n",
      "\tsampled: 746\n",
      "\ttp: 6175 fp: 128425 tn: 443885 fn: 1876\n",
      "threshold: 0.3\n",
      "\tsampled: 725\n",
      "\ttp: 6340 fp: 149267 tn: 411055 fn: 1489\n",
      "threshold: 0.25\n",
      "\tsampled: 766\n",
      "\ttp: 7111 fp: 172585 tn: 421250 fn: 1398\n",
      "threshold: 0.2\n",
      "\tsampled: 748\n",
      "\ttp: 6948 fp: 220144 tn: 361853 fn: 962\n",
      "threshold: 0.15\n",
      "\tsampled: 732\n",
      "\ttp: 6871 fp: 254204 tn: 298799 fn: 643\n",
      "threshold: 0.1\n",
      "\tsampled: 725\n",
      "\ttp: 7493 fp: 323612 tn: 227115 fn: 353\n",
      "threshold: 0.05\n",
      "\tsampled: 774\n",
      "\ttp: 7984 fp: 433570 tn: 151329 fn: 147\n",
      "threshold: 0.65\n",
      "\tsampled: 691\n",
      "\ttp: 4156 fp: 41281 tn: 495165 fn: 3290\n",
      "threshold: 0.55\n",
      "\tsampled: 727\n",
      "\ttp: 4931 fp: 61461 tn: 495311 fn: 2821\n",
      "threshold: 0.45\n",
      "\tsampled: 736\n",
      "\ttp: 5567 fp: 95329 tn: 467864 fn: 2494\n",
      "threshold: 0.4\n",
      "\tsampled: 712\n",
      "\ttp: 5767 fp: 95238 tn: 461299 fn: 1984\n",
      "threshold: 0.35\n",
      "\tsampled: 740\n",
      "\ttp: 6424 fp: 125145 tn: 423245 fn: 1822\n",
      "threshold: 0.3\n",
      "\tsampled: 663\n",
      "\ttp: 5749 fp: 140565 tn: 374348 fn: 1239\n",
      "threshold: 0.25\n",
      "\tsampled: 693\n",
      "\ttp: 6433 fp: 161141 tn: 367021 fn: 1096\n",
      "threshold: 0.2\n",
      "\tsampled: 772\n",
      "\ttp: 7412 fp: 212417 tn: 373901 fn: 1015\n",
      "threshold: 0.15\n",
      "\tsampled: 715\n",
      "\ttp: 7176 fp: 252544 tn: 294922 fn: 637\n",
      "threshold: 0.1\n",
      "\tsampled: 739\n",
      "\ttp: 7822 fp: 327854 tn: 238154 fn: 376\n",
      "threshold: 0.05\n",
      "\tsampled: 719\n",
      "\ttp: 7652 fp: 406957 tn: 146326 fn: 116\n",
      "threshold: 0.65\n",
      "\tsampled: 761\n",
      "\ttp: 4498 fp: 51667 tn: 552165 fn: 3799\n",
      "threshold: 0.55\n",
      "\tsampled: 731\n",
      "\ttp: 4979 fp: 62963 tn: 488098 fn: 2792\n",
      "threshold: 0.45\n",
      "\tsampled: 769\n",
      "\ttp: 5857 fp: 86999 tn: 493589 fn: 2440\n",
      "threshold: 0.4\n",
      "\tsampled: 728\n",
      "\ttp: 5632 fp: 104476 tn: 459510 fn: 2007\n",
      "threshold: 0.35\n",
      "\tsampled: 731\n",
      "\ttp: 6175 fp: 113506 tn: 439323 fn: 1739\n",
      "threshold: 0.3\n",
      "\tsampled: 776\n",
      "\ttp: 6958 fp: 154744 tn: 456319 fn: 1468\n",
      "threshold: 0.25\n",
      "\tsampled: 689\n",
      "\ttp: 6374 fp: 163180 tn: 369540 fn: 1083\n",
      "threshold: 0.2\n",
      "\tsampled: 772\n",
      "\ttp: 7417 fp: 219974 tn: 361660 fn: 950\n",
      "threshold: 0.15\n",
      "\tsampled: 729\n",
      "\ttp: 7416 fp: 262899 tn: 303595 fn: 712\n",
      "threshold: 0.1\n",
      "\tsampled: 754\n",
      "\ttp: 7736 fp: 339956 tn: 237757 fn: 307\n",
      "threshold: 0.05\n",
      "\tsampled: 758\n",
      "\ttp: 7995 fp: 428351 tn: 160385 fn: 154\n"
     ]
    }
   ],
   "source": [
    "# randomly sample a subset of the database\n",
    "np.random.seed(428)\n",
    "\n",
    "# test classification accuracy using complete database\n",
    "metrics = []\n",
    "for rep in range(3):\n",
    "    mets = {}\n",
    "    for threshold in [0.65, 0.55, 0.45, 0.4, 0.35, 0.3, 0.25, 0.2, 0.15, 0.1, 0.05]:\n",
    "        print('threshold:', threshold)\n",
    "        tp, fp, tn, fn = 0, 0, 0, 0\n",
    "        qry_sel1 = 'SELECT nistno, name, adduct, pre_mz, msms_mz, msms_i FROM msms;'\n",
    "        qry_sel2 = 'SELECT name, adduct, msms_mz, msms_i FROM msms WHERE pre_mz>=? AND pre_mz<=? AND nistno!=?;'\n",
    "        i = 0\n",
    "        for nistno, name, adduct, pre_mz, msms_mz, msms_i in cur_nist.execute(qry_sel1).fetchall():\n",
    "            if np.random.random() >= 1 / 500:  # sample around 1000 per threshold\n",
    "                # skip this iteration of the loop\n",
    "                continue\n",
    "            mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "            spectrum = spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                                  np.fromstring(msms_i.strip('[]'), sep=',')])\n",
    "            for name2, adduct2, msms_mz2, msms_i2 in cur_nist.execute(qry_sel2, (mz_min, mz_max, nistno)).fetchall():\n",
    "                spectrum2 = spec_norm([np.fromstring(msms_mz2.strip('[]'), sep=','), \n",
    "                                       np.fromstring(msms_i2.strip('[]'), sep=',')])\n",
    "                # label based on similarity threshold\n",
    "                label = spec_entropy_similarity(spectrum, spectrum2) >= threshold\n",
    "                # actual ground truth\n",
    "                known = (name == name2) and (adduct == adduct2)\n",
    "                # count TP, FP, TN, or FN\n",
    "                if label:\n",
    "                    if known:\n",
    "                        tp += 1\n",
    "                    else:\n",
    "                        fp += 1\n",
    "                else:\n",
    "                    if known:\n",
    "                        fn += 1\n",
    "                    else:\n",
    "                        tn += 1\n",
    "            i += 1\n",
    "        print('\\tsampled:', i)\n",
    "        print('\\ttp:', tp, 'fp:', fp, 'tn:', tn, 'fn:', fn)\n",
    "        mets[threshold] = {'sampled': i, 'tp': tp, 'fp': fp, 'tn': tn, 'fn': fn}\n",
    "    metrics.append(mets)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "52777592",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('exclude_target_match_test_2.json', 'w') as jf:\n",
    "    json.dump(metrics, jf, indent=4)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "536ed480",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "processed 1000 spectra groups\n",
      "processed 2000 spectra groups\n",
      "processed 3000 spectra groups\n",
      "processed 4000 spectra groups\n",
      "processed 5000 spectra groups\n",
      "processed 6000 spectra groups\n",
      "processed 7000 spectra groups\n",
      "processed 8000 spectra groups\n",
      "processed 9000 spectra groups\n",
      "processed 10000 spectra groups\n",
      "processed 11000 spectra groups\n",
      "processed 12000 spectra groups\n",
      "processed 13000 spectra groups\n",
      "processed 14000 spectra groups\n",
      "processed 15000 spectra groups\n",
      "processed 16000 spectra groups\n",
      "processed 17000 spectra groups\n",
      "processed 18000 spectra groups\n",
      "processed 19000 spectra groups\n",
      "processed 20000 spectra groups\n",
      "processed 21000 spectra groups\n",
      "processed 22000 spectra groups\n",
      "processed 23000 spectra groups\n",
      "processed 24000 spectra groups\n",
      "processed 25000 spectra groups\n",
      "processed 26000 spectra groups\n",
      "processed 27000 spectra groups\n",
      "processed 28000 spectra groups\n",
      "processed 29000 spectra groups\n",
      "processed 30000 spectra groups\n",
      "processed 31000 spectra groups\n",
      "processed 32000 spectra groups\n",
      "processed 33000 spectra groups\n",
      "processed 34000 spectra groups\n",
      "processed 35000 spectra groups\n",
      "processed 36000 spectra groups\n",
      "processed 37000 spectra groups\n",
      "processed 38000 spectra groups\n",
      "processed 39000 spectra groups\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "\n",
    "# create a database with the combined spectra\n",
    "nist_comb_db_f = 'nist20_subset_combined.db'\n",
    "if os.path.isfile(nist_comb_db_f):\n",
    "    os.remove(nist_comb_db_f)\n",
    "qry_create = \"\"\"\n",
    "    CREATE TABLE msms (\n",
    "        name TEXT NOT NULL,\n",
    "        adduct TEXT NOT NULL,\n",
    "        pre_mz REAL NOT NULL,\n",
    "        ces TEXT NOT NULL,\n",
    "        msms_mz BLOB NOT NULL,\n",
    "        msms_i BLOB NOT NULL,\n",
    "        mean_spec_sim REAL NOT NULL\n",
    "    );\n",
    "    \"\"\" \n",
    "con_nist_c = sqlite3.connect(nist_comb_db_f)\n",
    "cur_nist_c = con_nist_c.cursor()\n",
    "cur_nist_c.execute(qry_create)\n",
    "\n",
    "# iterate through database selecting all pairs of name, adduct\n",
    "# construct groups from all spectra with same name, adduct\n",
    "# store the combined spectra for searching against later\n",
    "qry_sel1 = 'SELECT name, adduct, pre_mz FROM msms GROUP BY name, adduct;'\n",
    "qry_sel2 = 'SELECT ce, msms_mz, msms_i FROM msms WHERE name=? AND adduct=?;'\n",
    "qry_ins = 'INSERT INTO msms VALUES (?,?,?,?,?,?,?);'\n",
    "i = 0\n",
    "for name, adduct, pre_mz in cur_nist.execute(qry_sel1).fetchall():\n",
    "    ces = []\n",
    "    spectra = []\n",
    "    j = 0\n",
    "    for ce, msms_mz, msms_i in cur_nist.execute(qry_sel2, (name, adduct)).fetchall():\n",
    "        ces.append(ce)\n",
    "        spectra.append(spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                                  np.fromstring(msms_i.strip('[]'), sep=',')]))\n",
    "        j += 1\n",
    "    cmz, ci = spec_combine(spectra)\n",
    "    mean_spec_sim = 1\n",
    "    if j > 1:\n",
    "        ssims = []\n",
    "        for spectrum in spectra:\n",
    "            ssims.append(spec_entropy_similarity(spectrum, [cmz, ci]))\n",
    "        mean_spec_sim = np.mean(ssims)\n",
    "    qdata = (name, adduct, pre_mz, str(ces), cmz.tobytes(), ci.tobytes(), mean_spec_sim)\n",
    "    cur_nist_c.execute(qry_ins, qdata)\n",
    "    i += 1\n",
    "    if i % 1000 == 0:\n",
    "        print('processed', i, 'spectra groups')     \n",
    "    \n",
    "con_nist_c.commit()\n",
    "con_nist_c.close()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "4ba5b9fd",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "processed 1000 spectra groups\n",
      "processed 2000 spectra groups\n",
      "processed 3000 spectra groups\n",
      "processed 4000 spectra groups\n",
      "processed 5000 spectra groups\n",
      "processed 6000 spectra groups\n",
      "processed 7000 spectra groups\n",
      "processed 8000 spectra groups\n",
      "processed 9000 spectra groups\n",
      "processed 10000 spectra groups\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "\n",
    "# create a database with the combined spectra\n",
    "mona_comb_db_f = 'MoNA_subset_combined.db'\n",
    "if os.path.isfile(mona_comb_db_f):\n",
    "    os.remove(mona_comb_db_f)\n",
    "qry_create = \"\"\"\n",
    "    CREATE TABLE msms (\n",
    "        name TEXT NOT NULL,\n",
    "        adduct TEXT NOT NULL,\n",
    "        pre_mz REAL NOT NULL,\n",
    "        ces TEXT NOT NULL,\n",
    "        msms_mz BLOB NOT NULL,\n",
    "        msms_i BLOB NOT NULL,\n",
    "        mean_spec_sim REAL NOT NULL\n",
    "    );\n",
    "    \"\"\" \n",
    "con_mona_c = sqlite3.connect(mona_comb_db_f)\n",
    "cur_mona_c = con_mona_c.cursor()\n",
    "cur_mona_c.execute(qry_create)\n",
    "\n",
    "# iterate through database selecting all pairs of name, adduct\n",
    "# construct groups from all spectra with same name, adduct\n",
    "# store the combined spectra for searching against later\n",
    "qry_sel1 = 'SELECT name, adduct, pre_mz FROM msms GROUP BY name, adduct;'\n",
    "qry_sel2 = 'SELECT ce, msms_mz, msms_i FROM msms WHERE name=? AND adduct=?;'\n",
    "qry_ins = 'INSERT INTO msms VALUES (?,?,?,?,?,?,?);'\n",
    "i = 0\n",
    "for name, adduct, pre_mz in cur_mona.execute(qry_sel1).fetchall():\n",
    "    ces = []\n",
    "    spectra = []\n",
    "    j = 0\n",
    "    for ce, msms_mz, msms_i in cur_mona.execute(qry_sel2, (name, adduct)).fetchall():\n",
    "        ces.append(ce)\n",
    "        spectra.append(spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                                  np.fromstring(msms_i.strip('[]'), sep=',')]))\n",
    "        j += 1\n",
    "    cmz, ci = spec_combine(spectra)\n",
    "    mean_spec_sim = 1\n",
    "    if j > 1:\n",
    "        ssims = []\n",
    "        for spectrum in spectra:\n",
    "            ssims.append(spec_entropy_similarity(spectrum, [cmz, ci]))\n",
    "        mean_spec_sim = np.mean(ssims)\n",
    "    qdata = (name, adduct, pre_mz, str(ces), cmz.tobytes(), ci.tobytes(), mean_spec_sim)\n",
    "    cur_mona_c.execute(qry_ins, qdata)\n",
    "    i += 1\n",
    "    if i % 1000 == 0:\n",
    "        print('processed', i, 'spectra groups')     \n",
    "    \n",
    "con_mona_c.commit()\n",
    "con_mona_c.close()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "id": "fce452c4",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "threshold: 0.05\n",
      "\tsampled: 984\n",
      "\ttp: 984 fp: 61571 tn: 13397 fn: 0\n",
      "threshold: 0.1\n",
      "\tsampled: 951\n",
      "\ttp: 951 fp: 52702 tn: 20690 fn: 0\n",
      "threshold: 0.15\n",
      "\tsampled: 1012\n",
      "\ttp: 1012 fp: 49424 tn: 29252 fn: 0\n",
      "threshold: 0.2\n",
      "\tsampled: 1036\n",
      "\ttp: 1036 fp: 42112 tn: 37130 fn: 0\n",
      "threshold: 0.25\n",
      "\tsampled: 1027\n",
      "\ttp: 1027 fp: 35394 tn: 42610 fn: 0\n",
      "threshold: 0.3\n",
      "\tsampled: 1003\n",
      "\ttp: 1003 fp: 28359 tn: 46009 fn: 0\n",
      "threshold: 0.35\n",
      "\tsampled: 930\n",
      "\ttp: 930 fp: 20763 tn: 49384 fn: 0\n",
      "threshold: 0.4\n",
      "\tsampled: 1010\n",
      "\ttp: 1008 fp: 17348 tn: 58985 fn: 2\n",
      "threshold: 0.45\n",
      "\tsampled: 1003\n",
      "\ttp: 1001 fp: 13034 tn: 62798 fn: 2\n",
      "threshold: 0.5\n",
      "\tsampled: 969\n",
      "\ttp: 958 fp: 8500 tn: 65223 fn: 11\n",
      "threshold: 0.55\n",
      "\tsampled: 991\n",
      "\ttp: 959 fp: 6160 tn: 70760 fn: 32\n",
      "threshold: 0.6\n",
      "\tsampled: 991\n",
      "\ttp: 908 fp: 4466 tn: 72484 fn: 83\n",
      "threshold: 0.65\n",
      "\tsampled: 970\n",
      "\ttp: 871 fp: 3132 tn: 68744 fn: 99\n",
      "threshold: 0.7\n",
      "\tsampled: 1031\n",
      "\ttp: 855 fp: 1952 tn: 76127 fn: 176\n",
      "threshold: 0.75\n",
      "\tsampled: 977\n",
      "\ttp: 741 fp: 1553 tn: 72669 fn: 236\n",
      "threshold: 0.8\n",
      "\tsampled: 970\n",
      "\ttp: 600 fp: 939 tn: 73453 fn: 370\n",
      "threshold: 0.85\n",
      "\tsampled: 1021\n",
      "\ttp: 454 fp: 624 tn: 74140 fn: 567\n",
      "threshold: 0.9\n",
      "\tsampled: 1054\n",
      "\ttp: 281 fp: 380 tn: 80643 fn: 773\n",
      "threshold: 0.95\n",
      "\tsampled: 986\n",
      "\ttp: 109 fp: 82 tn: 73758 fn: 877\n",
      "threshold: 0.05\n",
      "\tsampled: 992\n",
      "\ttp: 992 fp: 61972 tn: 13795 fn: 0\n",
      "threshold: 0.1\n",
      "\tsampled: 989\n",
      "\ttp: 989 fp: 54561 tn: 19748 fn: 0\n",
      "threshold: 0.15\n",
      "\tsampled: 949\n",
      "\ttp: 949 fp: 43215 tn: 27244 fn: 0\n",
      "threshold: 0.2\n",
      "\tsampled: 1040\n",
      "\ttp: 1040 fp: 42585 tn: 35102 fn: 0\n",
      "threshold: 0.25\n",
      "\tsampled: 1026\n",
      "\ttp: 1026 fp: 36733 tn: 40438 fn: 0\n",
      "threshold: 0.3\n",
      "\tsampled: 999\n",
      "\ttp: 999 fp: 30214 tn: 47310 fn: 0\n",
      "threshold: 0.35\n",
      "\tsampled: 1031\n",
      "\ttp: 1031 fp: 22379 tn: 54150 fn: 0\n",
      "threshold: 0.4\n",
      "\tsampled: 1022\n",
      "\ttp: 1022 fp: 17578 tn: 57882 fn: 0\n",
      "threshold: 0.45\n",
      "\tsampled: 996\n",
      "\ttp: 994 fp: 13837 tn: 61197 fn: 2\n",
      "threshold: 0.5\n",
      "\tsampled: 988\n",
      "\ttp: 974 fp: 8574 tn: 65826 fn: 14\n",
      "threshold: 0.55\n",
      "\tsampled: 1006\n",
      "\ttp: 984 fp: 7095 tn: 72204 fn: 22\n",
      "threshold: 0.6\n",
      "\tsampled: 985\n",
      "\ttp: 912 fp: 4005 tn: 70381 fn: 73\n",
      "threshold: 0.65\n",
      "\tsampled: 1032\n",
      "\ttp: 936 fp: 2975 tn: 75889 fn: 96\n",
      "threshold: 0.7\n",
      "\tsampled: 1003\n",
      "\ttp: 823 fp: 2241 tn: 72709 fn: 180\n",
      "threshold: 0.75\n",
      "\tsampled: 1008\n",
      "\ttp: 755 fp: 1399 tn: 72865 fn: 253\n",
      "threshold: 0.8\n",
      "\tsampled: 1029\n",
      "\ttp: 626 fp: 1045 tn: 76525 fn: 403\n",
      "threshold: 0.85\n",
      "\tsampled: 1029\n",
      "\ttp: 477 fp: 515 tn: 75169 fn: 552\n",
      "threshold: 0.9\n",
      "\tsampled: 993\n",
      "\ttp: 280 fp: 466 tn: 77718 fn: 713\n",
      "threshold: 0.95\n",
      "\tsampled: 943\n",
      "\ttp: 109 fp: 164 tn: 73419 fn: 834\n",
      "threshold: 0.05\n",
      "\tsampled: 983\n",
      "\ttp: 983 fp: 61303 tn: 12488 fn: 0\n",
      "threshold: 0.1\n",
      "\tsampled: 962\n",
      "\ttp: 962 fp: 51261 tn: 20569 fn: 0\n",
      "threshold: 0.15\n",
      "\tsampled: 988\n",
      "\ttp: 988 fp: 44473 tn: 28883 fn: 0\n",
      "threshold: 0.2\n",
      "\tsampled: 1045\n",
      "\ttp: 1045 fp: 42433 tn: 36239 fn: 0\n",
      "threshold: 0.25\n",
      "\tsampled: 997\n",
      "\ttp: 997 fp: 34247 tn: 39145 fn: 0\n",
      "threshold: 0.3\n",
      "\tsampled: 995\n",
      "\ttp: 995 fp: 30308 tn: 45538 fn: 0\n",
      "threshold: 0.35\n",
      "\tsampled: 1018\n",
      "\ttp: 1018 fp: 23779 tn: 54123 fn: 0\n",
      "threshold: 0.4\n",
      "\tsampled: 1005\n",
      "\ttp: 1004 fp: 17378 tn: 57269 fn: 1\n",
      "threshold: 0.45\n",
      "\tsampled: 1021\n",
      "\ttp: 1016 fp: 14603 tn: 66672 fn: 5\n",
      "threshold: 0.5\n",
      "\tsampled: 1063\n",
      "\ttp: 1045 fp: 9656 tn: 70406 fn: 18\n",
      "threshold: 0.55\n",
      "\tsampled: 965\n",
      "\ttp: 933 fp: 6096 tn: 66528 fn: 32\n",
      "threshold: 0.6\n",
      "\tsampled: 958\n",
      "\ttp: 904 fp: 4082 tn: 66795 fn: 54\n",
      "threshold: 0.65\n",
      "\tsampled: 1018\n",
      "\ttp: 895 fp: 2886 tn: 75442 fn: 123\n",
      "threshold: 0.7\n",
      "\tsampled: 896\n",
      "\ttp: 744 fp: 1832 tn: 66380 fn: 152\n",
      "threshold: 0.75\n",
      "\tsampled: 967\n",
      "\ttp: 699 fp: 1135 tn: 73593 fn: 268\n",
      "threshold: 0.8\n",
      "\tsampled: 1038\n",
      "\ttp: 625 fp: 1044 tn: 74186 fn: 413\n",
      "threshold: 0.85\n",
      "\tsampled: 997\n",
      "\ttp: 453 fp: 649 tn: 73131 fn: 544\n",
      "threshold: 0.9\n",
      "\tsampled: 959\n",
      "\ttp: 236 fp: 231 tn: 72396 fn: 723\n",
      "threshold: 0.95\n",
      "\tsampled: 1016\n",
      "\ttp: 135 fp: 168 tn: 78057 fn: 881\n"
     ]
    }
   ],
   "source": [
    "con_nist_c = sqlite3.connect(nist_comb_db_f)\n",
    "cur_nist_c = con_nist_c.cursor()\n",
    "\n",
    "# randomly sample a subset of the database\n",
    "np.random.seed(469)\n",
    "\n",
    "# test classification accuracy using complete database\n",
    "metrics = []\n",
    "for rep in range(3):\n",
    "    mets = {}\n",
    "    for threshold in np.arange(0.05, 1.0, 0.05):\n",
    "        threshold = round(threshold, 2)\n",
    "        print('threshold:', threshold)\n",
    "        tp, fp, tn, fn = 0, 0, 0, 0\n",
    "        qry_sel1 = 'SELECT name, adduct, pre_mz, msms_mz, msms_i FROM msms;'\n",
    "        qry_sel2 = 'SELECT name, adduct, msms_mz, msms_i FROM msms WHERE pre_mz>=? AND pre_mz<=?;'\n",
    "        i = 0\n",
    "        for name, adduct, pre_mz, msms_mz, msms_i in cur_nist.execute(qry_sel1).fetchall():\n",
    "            if np.random.random() >= 1 / 370:  # sample around 1000 per threshold\n",
    "                # skip this iteration of the loop\n",
    "                continue\n",
    "            mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "            spectrum = spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                                  np.fromstring(msms_i.strip('[]'), sep=',')])\n",
    "            for name2, adduct2, msms_mz2, msms_i2 in cur_nist_c.execute(qry_sel2, (mz_min, mz_max)).fetchall():\n",
    "                spectrum2 = [np.frombuffer(msms_mz2), np.frombuffer(msms_i2)]\n",
    "                # label based on similarity threshold\n",
    "                label = spec_entropy_similarity(spectrum, spectrum2) >= threshold\n",
    "                # actual ground truth\n",
    "                known = (name == name2) and (adduct == adduct2)\n",
    "                # count TP, FP, TN, or FN\n",
    "                if label:\n",
    "                    if known:\n",
    "                        tp += 1\n",
    "                    else:\n",
    "                        fp += 1\n",
    "                else:\n",
    "                    if known:\n",
    "                        fn += 1\n",
    "                    else:\n",
    "                        tn += 1\n",
    "            i += 1\n",
    "        print('\\tsampled:', i)\n",
    "        print('\\ttp:', tp, 'fp:', fp, 'tn:', tn, 'fn:', fn)\n",
    "        mets[threshold] = {'sampled': i, 'tp': tp, 'fp': fp, 'tn': tn, 'fn': fn}\n",
    "    metrics.append(mets)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "id": "4e8efb93",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('combined_spectra_match_test.json', 'w') as jf:\n",
    "    json.dump(metrics, jf, indent=4)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "id": "44bcbbf5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "threshold: 0.05\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/93/rxb_7lxj5dnfzlgq40hwpl254g3nd8/T/ipykernel_16257/693147320.py:10: RuntimeWarning: divide by zero encountered in log\n",
      "  return -np.sum(i * np.log(i))\n",
      "/var/folders/93/rxb_7lxj5dnfzlgq40hwpl254g3nd8/T/ipykernel_16257/693147320.py:10: RuntimeWarning: invalid value encountered in multiply\n",
      "  return -np.sum(i * np.log(i))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tsampled: 2047\n",
      "\ttp: 30059 fp: 107333 tn: 85792 fn: 3829\n",
      "threshold: 0.1\n",
      "\tsampled: 2014\n",
      "\ttp: 29781 fp: 82727 tn: 107297 fn: 5488\n",
      "threshold: 0.15\n",
      "\tsampled: 1973\n",
      "\ttp: 26446 fp: 65649 tn: 119715 fn: 6149\n",
      "threshold: 0.2\n",
      "\tsampled: 2031\n",
      "\ttp: 26942 fp: 54577 tn: 132638 fn: 8472\n",
      "threshold: 0.25\n",
      "\tsampled: 2116\n",
      "\ttp: 26446 fp: 46957 tn: 147866 fn: 9158\n",
      "threshold: 0.3\n",
      "\tsampled: 2080\n",
      "\ttp: 24316 fp: 40977 tn: 150589 fn: 10263\n",
      "threshold: 0.35\n",
      "\tsampled: 1962\n",
      "\ttp: 22560 fp: 34048 tn: 147820 fn: 10197\n",
      "threshold: 0.4\n",
      "\tsampled: 2093\n",
      "\ttp: 22970 fp: 30294 tn: 167083 fn: 12871\n",
      "threshold: 0.45\n",
      "\tsampled: 2067\n",
      "\ttp: 20369 fp: 24873 tn: 167453 fn: 12520\n",
      "threshold: 0.5\n",
      "\tsampled: 2074\n",
      "\ttp: 20084 fp: 23664 tn: 173106 fn: 14050\n",
      "threshold: 0.55\n",
      "\tsampled: 2085\n",
      "\ttp: 19860 fp: 20204 tn: 174055 fn: 15244\n",
      "threshold: 0.6\n",
      "\tsampled: 2094\n",
      "\ttp: 18333 fp: 18292 tn: 177722 fn: 17171\n",
      "threshold: 0.65\n",
      "\tsampled: 2080\n",
      "\ttp: 16949 fp: 16718 tn: 187389 fn: 16862\n",
      "threshold: 0.7\n",
      "\tsampled: 1960\n",
      "\ttp: 14675 fp: 13187 tn: 171995 fn: 17742\n",
      "threshold: 0.75\n",
      "\tsampled: 2012\n",
      "\ttp: 14000 fp: 11242 tn: 176681 fn: 20936\n",
      "threshold: 0.8\n",
      "\tsampled: 2056\n",
      "\ttp: 13118 fp: 8698 tn: 183892 fn: 22405\n",
      "threshold: 0.85\n",
      "\tsampled: 2106\n",
      "\ttp: 11014 fp: 7674 tn: 192965 fn: 23257\n",
      "threshold: 0.9\n",
      "\tsampled: 2022\n",
      "\ttp: 8501 fp: 5228 tn: 184859 fn: 23179\n",
      "threshold: 0.95\n",
      "\tsampled: 2035\n",
      "\ttp: 6800 fp: 2989 tn: 183331 fn: 27459\n",
      "threshold: 0.05\n",
      "\tsampled: 2075\n",
      "\ttp: 30773 fp: 110514 tn: 88219 fn: 4050\n",
      "threshold: 0.1\n",
      "\tsampled: 2095\n",
      "\ttp: 28489 fp: 83613 tn: 106690 fn: 5323\n",
      "threshold: 0.15\n",
      "\tsampled: 1947\n",
      "\ttp: 24773 fp: 64343 tn: 113689 fn: 5943\n",
      "threshold: 0.2\n",
      "\tsampled: 2016\n",
      "\ttp: 25532 fp: 53585 tn: 135117 fn: 8357\n",
      "threshold: 0.25\n",
      "\tsampled: 2090\n",
      "\ttp: 26577 fp: 46683 tn: 146857 fn: 9461\n",
      "threshold: 0.3\n",
      "\tsampled: 2092\n",
      "\ttp: 23772 fp: 40912 tn: 152943 fn: 9214\n",
      "threshold: 0.35\n",
      "\tsampled: 2065\n",
      "\ttp: 23378 fp: 35154 tn: 157310 fn: 11956\n",
      "threshold: 0.4\n",
      "\tsampled: 1983\n",
      "\ttp: 22153 fp: 29543 tn: 158121 fn: 12041\n",
      "threshold: 0.45\n",
      "\tsampled: 2054\n",
      "\ttp: 20433 fp: 27358 tn: 171839 fn: 12485\n",
      "threshold: 0.5\n",
      "\tsampled: 1978\n",
      "\ttp: 19904 fp: 22918 tn: 164441 fn: 13101\n",
      "threshold: 0.55\n",
      "\tsampled: 1994\n",
      "\ttp: 18697 fp: 19913 tn: 161736 fn: 14208\n",
      "threshold: 0.6\n",
      "\tsampled: 2033\n",
      "\ttp: 17802 fp: 17616 tn: 176682 fn: 15310\n",
      "threshold: 0.65\n",
      "\tsampled: 2075\n",
      "\ttp: 16694 fp: 13825 tn: 179072 fn: 17719\n",
      "threshold: 0.7\n",
      "\tsampled: 2009\n",
      "\ttp: 15214 fp: 12985 tn: 175883 fn: 18503\n",
      "threshold: 0.75\n",
      "\tsampled: 2080\n",
      "\ttp: 15110 fp: 11545 tn: 189362 fn: 21810\n",
      "threshold: 0.8\n",
      "\tsampled: 2033\n",
      "\ttp: 13192 fp: 9328 tn: 184197 fn: 24343\n",
      "threshold: 0.85\n",
      "\tsampled: 2069\n",
      "\ttp: 10282 fp: 7266 tn: 184931 fn: 23576\n",
      "threshold: 0.9\n",
      "\tsampled: 2108\n",
      "\ttp: 9202 fp: 5627 tn: 188670 fn: 24926\n",
      "threshold: 0.95\n",
      "\tsampled: 2088\n",
      "\ttp: 7026 fp: 3136 tn: 197769 fn: 28640\n",
      "threshold: 0.05\n",
      "\tsampled: 2081\n",
      "\ttp: 28775 fp: 104666 tn: 85126 fn: 3544\n",
      "threshold: 0.1\n",
      "\tsampled: 2001\n",
      "\ttp: 27805 fp: 80621 tn: 105572 fn: 5534\n",
      "threshold: 0.15\n",
      "\tsampled: 2091\n",
      "\ttp: 27799 fp: 71170 tn: 124892 fn: 6796\n",
      "threshold: 0.2\n",
      "\tsampled: 2049\n",
      "\ttp: 25303 fp: 53487 tn: 134602 fn: 7680\n",
      "threshold: 0.25\n",
      "\tsampled: 2090\n",
      "\ttp: 25612 fp: 48141 tn: 147634 fn: 9354\n",
      "threshold: 0.3\n",
      "\tsampled: 2077\n",
      "\ttp: 24794 fp: 43115 tn: 157091 fn: 9926\n",
      "threshold: 0.35\n",
      "\tsampled: 2011\n",
      "\ttp: 22842 fp: 36200 tn: 153808 fn: 9675\n",
      "threshold: 0.4\n",
      "\tsampled: 2017\n",
      "\ttp: 22038 fp: 29323 tn: 163567 fn: 12156\n",
      "threshold: 0.45\n",
      "\tsampled: 2144\n",
      "\ttp: 22603 fp: 28782 tn: 170988 fn: 14271\n",
      "threshold: 0.5\n",
      "\tsampled: 2053\n",
      "\ttp: 19726 fp: 22537 tn: 170913 fn: 13356\n",
      "threshold: 0.55\n",
      "\tsampled: 2020\n",
      "\ttp: 18931 fp: 20285 tn: 171756 fn: 14560\n",
      "threshold: 0.6\n",
      "\tsampled: 2022\n",
      "\ttp: 17048 fp: 17674 tn: 169488 fn: 15534\n",
      "threshold: 0.65\n",
      "\tsampled: 2029\n",
      "\ttp: 16609 fp: 14841 tn: 182497 fn: 17342\n",
      "threshold: 0.7\n",
      "\tsampled: 2095\n",
      "\ttp: 14792 fp: 13859 tn: 176510 fn: 19844\n",
      "threshold: 0.75\n",
      "\tsampled: 2011\n",
      "\ttp: 13305 fp: 10916 tn: 172749 fn: 19069\n",
      "threshold: 0.8\n",
      "\tsampled: 2101\n",
      "\ttp: 13269 fp: 9830 tn: 186532 fn: 22859\n",
      "threshold: 0.85\n",
      "\tsampled: 2056\n",
      "\ttp: 11811 fp: 6498 tn: 187474 fn: 24640\n",
      "threshold: 0.9\n",
      "\tsampled: 1952\n",
      "\ttp: 8475 fp: 4884 tn: 180608 fn: 23611\n",
      "threshold: 0.95\n",
      "\tsampled: 2162\n",
      "\ttp: 7193 fp: 3526 tn: 200795 fn: 27817\n"
     ]
    }
   ],
   "source": [
    "# randomly sample a subset of the database\n",
    "np.random.seed(469)\n",
    "\n",
    "# test classification accuracy using complete database\n",
    "metrics = []\n",
    "for rep in range(3):\n",
    "    mets = {}\n",
    "    for threshold in np.arange(0.05, 1.0, 0.05):\n",
    "        threshold = round(threshold, 2)\n",
    "        print('threshold:', threshold)\n",
    "        tp, fp, tn, fn = 0, 0, 0, 0\n",
    "        qry_sel1 = 'SELECT name, adduct, pre_mz, msms_mz, msms_i FROM msms;'\n",
    "        qry_sel2 = 'SELECT name, adduct, msms_mz, msms_i FROM msms WHERE pre_mz>=? AND pre_mz<=?;'\n",
    "        i = 0\n",
    "        for name, adduct, pre_mz, msms_mz, msms_i in cur_mona.execute(qry_sel1).fetchall():\n",
    "            if np.random.random() >= 1 / 25:  # sample around 2000 per threshold\n",
    "                # skip this iteration of the loop\n",
    "                continue\n",
    "            mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "            spectrum = spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                                  np.fromstring(msms_i.strip('[]'), sep=',')])\n",
    "            for name2, adduct2, msms_mz2, msms_i2 in cur_mona.execute(qry_sel2, (mz_min, mz_max)).fetchall():\n",
    "                spectrum2 = spec_norm([np.fromstring(msms_mz2.strip('[]'), sep=','), \n",
    "                                       np.fromstring(msms_i2.strip('[]'), sep=',')])\n",
    "                # label based on similarity threshold\n",
    "                label = spec_entropy_similarity(spectrum, spectrum2) >= threshold\n",
    "                # actual ground truth\n",
    "                known = (name == name2) and (adduct == adduct2)\n",
    "                # count TP, FP, TN, or FN\n",
    "                if label:\n",
    "                    if known:\n",
    "                        tp += 1\n",
    "                    else:\n",
    "                        fp += 1\n",
    "                else:\n",
    "                    if known:\n",
    "                        fn += 1\n",
    "                    else:\n",
    "                        tn += 1\n",
    "            i += 1\n",
    "        print('\\tsampled:', i)\n",
    "        print('\\ttp:', tp, 'fp:', fp, 'tn:', tn, 'fn:', fn)\n",
    "        mets[threshold] = {'sampled': i, 'tp': tp, 'fp': fp, 'tn': tn, 'fn': fn}\n",
    "    metrics.append(mets)\n",
    "\n",
    "    \n",
    "with open('full_match_test.json', 'w') as jf:\n",
    "    json.dump(metrics, jf, indent=4)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "id": "2deb6e62",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "threshold: 0.05\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/93/rxb_7lxj5dnfzlgq40hwpl254g3nd8/T/ipykernel_16257/693147320.py:10: RuntimeWarning: divide by zero encountered in log\n",
      "  return -np.sum(i * np.log(i))\n",
      "/var/folders/93/rxb_7lxj5dnfzlgq40hwpl254g3nd8/T/ipykernel_16257/693147320.py:10: RuntimeWarning: invalid value encountered in multiply\n",
      "  return -np.sum(i * np.log(i))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tsampled: 2047\n",
      "\ttp: 28012 fp: 107333 tn: 85792 fn: 3829\n",
      "threshold: 0.1\n",
      "\tsampled: 2014\n",
      "\ttp: 27770 fp: 82727 tn: 107297 fn: 5485\n",
      "threshold: 0.15\n",
      "\tsampled: 1973\n",
      "\ttp: 24477 fp: 65649 tn: 119715 fn: 6145\n",
      "threshold: 0.2\n",
      "\tsampled: 2031\n",
      "\ttp: 24917 fp: 54577 tn: 132638 fn: 8466\n",
      "threshold: 0.25\n",
      "\tsampled: 2116\n",
      "\ttp: 24334 fp: 46957 tn: 147866 fn: 9154\n",
      "threshold: 0.3\n",
      "\tsampled: 2080\n",
      "\ttp: 22236 fp: 40977 tn: 150589 fn: 10263\n",
      "threshold: 0.35\n",
      "\tsampled: 1962\n",
      "\ttp: 20599 fp: 34048 tn: 147820 fn: 10196\n",
      "threshold: 0.4\n",
      "\tsampled: 2093\n",
      "\ttp: 20879 fp: 30294 tn: 167083 fn: 12869\n",
      "threshold: 0.45\n",
      "\tsampled: 2067\n",
      "\ttp: 18307 fp: 24873 tn: 167453 fn: 12515\n",
      "threshold: 0.5\n",
      "\tsampled: 2074\n",
      "\ttp: 18013 fp: 23664 tn: 173106 fn: 14047\n",
      "threshold: 0.55\n",
      "\tsampled: 2085\n",
      "\ttp: 17778 fp: 20204 tn: 174055 fn: 15241\n",
      "threshold: 0.6\n",
      "\tsampled: 2094\n",
      "\ttp: 16240 fp: 18292 tn: 177722 fn: 17170\n",
      "threshold: 0.65\n",
      "\tsampled: 2080\n",
      "\ttp: 14872 fp: 16718 tn: 187389 fn: 16859\n",
      "threshold: 0.7\n",
      "\tsampled: 1960\n",
      "\ttp: 12719 fp: 13187 tn: 171995 fn: 17738\n",
      "threshold: 0.75\n",
      "\tsampled: 2012\n",
      "\ttp: 11991 fp: 11242 tn: 176681 fn: 20933\n",
      "threshold: 0.8\n",
      "\tsampled: 2056\n",
      "\ttp: 11068 fp: 8698 tn: 183892 fn: 22399\n",
      "threshold: 0.85\n",
      "\tsampled: 2106\n",
      "\ttp: 8908 fp: 7674 tn: 192965 fn: 23257\n",
      "threshold: 0.9\n",
      "\tsampled: 2022\n",
      "\ttp: 6483 fp: 5228 tn: 184859 fn: 23175\n",
      "threshold: 0.95\n",
      "\tsampled: 2035\n",
      "\ttp: 4769 fp: 2989 tn: 183331 fn: 27455\n",
      "threshold: 0.05\n",
      "\tsampled: 2075\n",
      "\ttp: 28700 fp: 110514 tn: 88219 fn: 4048\n",
      "threshold: 0.1\n",
      "\tsampled: 2095\n",
      "\ttp: 26394 fp: 83613 tn: 106690 fn: 5323\n",
      "threshold: 0.15\n",
      "\tsampled: 1947\n",
      "\ttp: 22826 fp: 64343 tn: 113689 fn: 5943\n",
      "threshold: 0.2\n",
      "\tsampled: 2016\n",
      "\ttp: 23517 fp: 53585 tn: 135117 fn: 8356\n",
      "threshold: 0.25\n",
      "\tsampled: 2090\n",
      "\ttp: 24489 fp: 46683 tn: 146857 fn: 9459\n",
      "threshold: 0.3\n",
      "\tsampled: 2092\n",
      "\ttp: 21683 fp: 40912 tn: 152943 fn: 9211\n",
      "threshold: 0.35\n",
      "\tsampled: 2065\n",
      "\ttp: 21318 fp: 35154 tn: 157310 fn: 11951\n",
      "threshold: 0.4\n",
      "\tsampled: 1983\n",
      "\ttp: 20172 fp: 29543 tn: 158121 fn: 12039\n",
      "threshold: 0.45\n",
      "\tsampled: 2054\n",
      "\ttp: 18382 fp: 27358 tn: 171839 fn: 12482\n",
      "threshold: 0.5\n",
      "\tsampled: 1978\n",
      "\ttp: 17928 fp: 22918 tn: 164441 fn: 13099\n",
      "threshold: 0.55\n",
      "\tsampled: 1994\n",
      "\ttp: 16706 fp: 19913 tn: 161736 fn: 14205\n",
      "threshold: 0.6\n",
      "\tsampled: 2033\n",
      "\ttp: 15772 fp: 17616 tn: 176682 fn: 15307\n",
      "threshold: 0.65\n",
      "\tsampled: 2075\n",
      "\ttp: 14623 fp: 13825 tn: 179072 fn: 17715\n",
      "threshold: 0.7\n",
      "\tsampled: 2009\n",
      "\ttp: 13209 fp: 12985 tn: 175883 fn: 18499\n",
      "threshold: 0.75\n",
      "\tsampled: 2080\n",
      "\ttp: 13031 fp: 11545 tn: 189362 fn: 21809\n",
      "threshold: 0.8\n",
      "\tsampled: 2033\n",
      "\ttp: 11162 fp: 9328 tn: 184197 fn: 24340\n",
      "threshold: 0.85\n",
      "\tsampled: 2069\n",
      "\ttp: 8218 fp: 7266 tn: 184931 fn: 23571\n",
      "threshold: 0.9\n",
      "\tsampled: 2108\n",
      "\ttp: 7099 fp: 5627 tn: 188670 fn: 24921\n",
      "threshold: 0.95\n",
      "\tsampled: 2088\n",
      "\ttp: 4947 fp: 3136 tn: 197769 fn: 28631\n",
      "threshold: 0.05\n",
      "\tsampled: 2081\n",
      "\ttp: 26697 fp: 104666 tn: 85126 fn: 3541\n",
      "threshold: 0.1\n",
      "\tsampled: 2001\n",
      "\ttp: 25804 fp: 80621 tn: 105572 fn: 5534\n",
      "threshold: 0.15\n",
      "\tsampled: 2091\n",
      "\ttp: 25710 fp: 71170 tn: 124892 fn: 6794\n",
      "threshold: 0.2\n",
      "\tsampled: 2049\n",
      "\ttp: 23255 fp: 53487 tn: 134602 fn: 7679\n",
      "threshold: 0.25\n",
      "\tsampled: 2090\n",
      "\ttp: 23523 fp: 48141 tn: 147634 fn: 9353\n",
      "threshold: 0.3\n",
      "\tsampled: 2077\n",
      "\ttp: 22722 fp: 43115 tn: 157091 fn: 9921\n",
      "threshold: 0.35\n",
      "\tsampled: 2011\n",
      "\ttp: 20834 fp: 36200 tn: 153808 fn: 9672\n",
      "threshold: 0.4\n",
      "\tsampled: 2017\n",
      "\ttp: 20023 fp: 29323 tn: 163567 fn: 12154\n",
      "threshold: 0.45\n",
      "\tsampled: 2144\n",
      "\ttp: 20466 fp: 28782 tn: 170988 fn: 14264\n",
      "threshold: 0.5\n",
      "\tsampled: 2053\n",
      "\ttp: 17676 fp: 22537 tn: 170913 fn: 13353\n",
      "threshold: 0.55\n",
      "\tsampled: 2020\n",
      "\ttp: 16911 fp: 20285 tn: 171756 fn: 14560\n",
      "threshold: 0.6\n",
      "\tsampled: 2022\n",
      "\ttp: 15029 fp: 17674 tn: 169488 fn: 15531\n",
      "threshold: 0.65\n",
      "\tsampled: 2029\n",
      "\ttp: 14584 fp: 14841 tn: 182497 fn: 17338\n",
      "threshold: 0.7\n",
      "\tsampled: 2095\n",
      "\ttp: 12699 fp: 13859 tn: 176510 fn: 19842\n",
      "threshold: 0.75\n",
      "\tsampled: 2011\n",
      "\ttp: 11296 fp: 10916 tn: 172749 fn: 19067\n",
      "threshold: 0.8\n",
      "\tsampled: 2101\n",
      "\ttp: 11171 fp: 9830 tn: 186532 fn: 22856\n",
      "threshold: 0.85\n",
      "\tsampled: 2056\n",
      "\ttp: 9759 fp: 6498 tn: 187474 fn: 24636\n",
      "threshold: 0.9\n",
      "\tsampled: 1952\n",
      "\ttp: 6529 fp: 4884 tn: 180608 fn: 23605\n",
      "threshold: 0.95\n",
      "\tsampled: 2162\n",
      "\ttp: 5037 fp: 3526 tn: 200795 fn: 27811\n"
     ]
    }
   ],
   "source": [
    "# randomly sample a subset of the database\n",
    "np.random.seed(469)\n",
    "\n",
    "# test classification accuracy using complete database\n",
    "metrics = []\n",
    "for rep in range(3):\n",
    "    mets = {}\n",
    "    for threshold in np.arange(0.05, 1.0, 0.05):\n",
    "        threshold = round(threshold, 2)\n",
    "        print('threshold:', threshold)\n",
    "        tp, fp, tn, fn = 0, 0, 0, 0\n",
    "        qry_sel1 = 'SELECT mona_id, name, adduct, pre_mz, msms_mz, msms_i FROM msms;'\n",
    "        qry_sel2 = 'SELECT name, adduct, msms_mz, msms_i FROM msms WHERE pre_mz>=? AND pre_mz<=? AND mona_id!=?;'\n",
    "        i = 0\n",
    "        for mona_id, name, adduct, pre_mz, msms_mz, msms_i in cur_mona.execute(qry_sel1).fetchall():\n",
    "            if np.random.random() >= 1 / 25:  # sample around 2000 per threshold\n",
    "                # skip this iteration of the loop\n",
    "                continue\n",
    "            mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "            spectrum = spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                                  np.fromstring(msms_i.strip('[]'), sep=',')])\n",
    "            for name2, adduct2, msms_mz2, msms_i2 in cur_mona.execute(qry_sel2, (mz_min, mz_max, mona_id)).fetchall():\n",
    "                spectrum2 = spec_norm([np.fromstring(msms_mz2.strip('[]'), sep=','), \n",
    "                                       np.fromstring(msms_i2.strip('[]'), sep=',')])\n",
    "                # label based on similarity threshold\n",
    "                label = spec_entropy_similarity(spectrum, spectrum2) >= threshold\n",
    "                # actual ground truth\n",
    "                known = (name == name2) and (adduct == adduct2)\n",
    "                # count TP, FP, TN, or FN\n",
    "                if label:\n",
    "                    if known:\n",
    "                        tp += 1\n",
    "                    else:\n",
    "                        fp += 1\n",
    "                else:\n",
    "                    if known:\n",
    "                        fn += 1\n",
    "                    else:\n",
    "                        tn += 1\n",
    "            i += 1\n",
    "        print('\\tsampled:', i)\n",
    "        print('\\ttp:', tp, 'fp:', fp, 'tn:', tn, 'fn:', fn)\n",
    "        mets[threshold] = {'sampled': i, 'tp': tp, 'fp': fp, 'tn': tn, 'fn': fn}\n",
    "    metrics.append(mets)\n",
    "\n",
    "    \n",
    "with open('mona_exclude_target_match_test.json', 'w') as jf:\n",
    "    json.dump(metrics, jf, indent=4)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "22384798",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "threshold: 0.05\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/93/rxb_7lxj5dnfzlgq40hwpl254g3nd8/T/ipykernel_98782/693147320.py:10: RuntimeWarning: divide by zero encountered in log\n",
      "  return -np.sum(i * np.log(i))\n",
      "/var/folders/93/rxb_7lxj5dnfzlgq40hwpl254g3nd8/T/ipykernel_98782/693147320.py:10: RuntimeWarning: invalid value encountered in multiply\n",
      "  return -np.sum(i * np.log(i))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\tsampled: 2047\n",
      "\ttp: 1939 fp: 21873 tn: 14586 fn: 9\n",
      "threshold: 0.1\n",
      "\tsampled: 2014\n",
      "\ttp: 1904 fp: 17183 tn: 18934 fn: 10\n",
      "threshold: 0.15\n",
      "\tsampled: 1973\n",
      "\ttp: 1875 fp: 14245 tn: 21071 fn: 9\n",
      "threshold: 0.2\n",
      "\tsampled: 2031\n",
      "\ttp: 1909 fp: 12161 tn: 23426 fn: 20\n",
      "threshold: 0.25\n",
      "\tsampled: 2116\n",
      "\ttp: 1992 fp: 10788 tn: 26870 fn: 18\n",
      "threshold: 0.3\n",
      "\tsampled: 2080\n",
      "\ttp: 1952 fp: 9151 tn: 27241 fn: 20\n",
      "threshold: 0.35\n",
      "\tsampled: 1962\n",
      "\ttp: 1836 fp: 7475 tn: 27062 fn: 30\n",
      "threshold: 0.4\n",
      "\tsampled: 2093\n",
      "\ttp: 1941 fp: 6368 tn: 31074 fn: 53\n",
      "threshold: 0.45\n",
      "\tsampled: 2067\n",
      "\ttp: 1870 fp: 5176 tn: 30911 fn: 99\n",
      "threshold: 0.5\n",
      "\tsampled: 2074\n",
      "\ttp: 1836 fp: 4739 tn: 32483 fn: 127\n",
      "threshold: 0.55\n",
      "\tsampled: 2085\n",
      "\ttp: 1766 fp: 4105 tn: 32872 fn: 206\n",
      "threshold: 0.6\n",
      "\tsampled: 2094\n",
      "\ttp: 1701 fp: 3590 tn: 34090 fn: 275\n",
      "threshold: 0.65\n",
      "\tsampled: 2080\n",
      "\ttp: 1573 fp: 2818 tn: 35309 fn: 381\n",
      "threshold: 0.7\n",
      "\tsampled: 1960\n",
      "\ttp: 1332 fp: 2116 tn: 32985 fn: 537\n",
      "threshold: 0.75\n",
      "\tsampled: 2012\n",
      "\ttp: 1199 fp: 1778 tn: 33846 fn: 723\n",
      "threshold: 0.8\n",
      "\tsampled: 2056\n",
      "\ttp: 1016 fp: 1355 tn: 35866 fn: 942\n",
      "threshold: 0.85\n",
      "\tsampled: 2106\n",
      "\ttp: 822 fp: 1048 tn: 36852 fn: 1178\n",
      "threshold: 0.9\n",
      "\tsampled: 2022\n",
      "\ttp: 586 fp: 605 tn: 36202 fn: 1334\n",
      "threshold: 0.95\n",
      "\tsampled: 2035\n",
      "\ttp: 419 fp: 336 tn: 35319 fn: 1519\n",
      "threshold: 0.05\n",
      "\tsampled: 2075\n",
      "\ttp: 1971 fp: 22159 tn: 14792 fn: 10\n",
      "threshold: 0.1\n",
      "\tsampled: 2095\n",
      "\ttp: 1979 fp: 17472 tn: 18908 fn: 11\n",
      "threshold: 0.15\n",
      "\tsampled: 1947\n",
      "\ttp: 1848 fp: 13556 tn: 20587 fn: 8\n",
      "threshold: 0.2\n",
      "\tsampled: 2016\n",
      "\ttp: 1902 fp: 12116 tn: 23786 fn: 14\n",
      "threshold: 0.25\n",
      "\tsampled: 2090\n",
      "\ttp: 1984 fp: 10161 tn: 26421 fn: 14\n",
      "threshold: 0.3\n",
      "\tsampled: 2092\n",
      "\ttp: 1968 fp: 9175 tn: 27644 fn: 21\n",
      "threshold: 0.35\n",
      "\tsampled: 2065\n",
      "\ttp: 1922 fp: 7864 tn: 29503 fn: 46\n",
      "threshold: 0.4\n",
      "\tsampled: 1983\n",
      "\ttp: 1833 fp: 6335 tn: 28768 fn: 44\n",
      "threshold: 0.45\n",
      "\tsampled: 2054\n",
      "\ttp: 1871 fp: 5727 tn: 31947 fn: 91\n",
      "threshold: 0.5\n",
      "\tsampled: 1978\n",
      "\ttp: 1779 fp: 4631 tn: 31303 fn: 108\n",
      "threshold: 0.55\n",
      "\tsampled: 1994\n",
      "\ttp: 1718 fp: 3949 tn: 30726 fn: 180\n",
      "threshold: 0.6\n",
      "\tsampled: 2033\n",
      "\ttp: 1694 fp: 3404 tn: 34054 fn: 244\n",
      "threshold: 0.65\n",
      "\tsampled: 2075\n",
      "\ttp: 1576 fp: 2496 tn: 34128 fn: 393\n",
      "threshold: 0.7\n",
      "\tsampled: 2009\n",
      "\ttp: 1355 fp: 2116 tn: 33643 fn: 547\n",
      "threshold: 0.75\n",
      "\tsampled: 2080\n",
      "\ttp: 1242 fp: 1808 tn: 36135 fn: 754\n",
      "threshold: 0.8\n",
      "\tsampled: 2033\n",
      "\ttp: 969 fp: 1386 tn: 34428 fn: 965\n",
      "threshold: 0.85\n",
      "\tsampled: 2069\n",
      "\ttp: 784 fp: 982 tn: 35745 fn: 1176\n",
      "threshold: 0.9\n",
      "\tsampled: 2108\n",
      "\ttp: 594 fp: 622 tn: 36598 fn: 1380\n",
      "threshold: 0.95\n",
      "\tsampled: 2088\n",
      "\ttp: 404 fp: 339 tn: 36835 fn: 1581\n",
      "threshold: 0.05\n",
      "\tsampled: 2081\n",
      "\ttp: 1954 fp: 21600 tn: 14528 fn: 12\n",
      "threshold: 0.1\n",
      "\tsampled: 2001\n",
      "\ttp: 1895 fp: 16884 tn: 18606 fn: 6\n",
      "threshold: 0.15\n",
      "\tsampled: 2091\n",
      "\ttp: 1979 fp: 14737 tn: 22086 fn: 10\n",
      "threshold: 0.2\n",
      "\tsampled: 2049\n",
      "\ttp: 1933 fp: 11987 tn: 24176 fn: 9\n",
      "threshold: 0.25\n",
      "\tsampled: 2090\n",
      "\ttp: 1961 fp: 10548 tn: 27114 fn: 19\n",
      "threshold: 0.3\n",
      "\tsampled: 2077\n",
      "\ttp: 1957 fp: 9334 tn: 27710 fn: 24\n",
      "threshold: 0.35\n",
      "\tsampled: 2011\n",
      "\ttp: 1881 fp: 7892 tn: 28395 fn: 34\n",
      "threshold: 0.4\n",
      "\tsampled: 2017\n",
      "\ttp: 1861 fp: 6225 tn: 29338 fn: 56\n",
      "threshold: 0.45\n",
      "\tsampled: 2144\n",
      "\ttp: 1955 fp: 6076 tn: 31837 fn: 94\n",
      "threshold: 0.5\n",
      "\tsampled: 2053\n",
      "\ttp: 1846 fp: 4529 tn: 31505 fn: 118\n",
      "threshold: 0.55\n",
      "\tsampled: 2020\n",
      "\ttp: 1732 fp: 3749 tn: 32001 fn: 180\n",
      "threshold: 0.6\n",
      "\tsampled: 2022\n",
      "\ttp: 1641 fp: 3242 tn: 32576 fn: 279\n",
      "threshold: 0.65\n",
      "\tsampled: 2029\n",
      "\ttp: 1520 fp: 2578 tn: 34568 fn: 389\n",
      "threshold: 0.7\n",
      "\tsampled: 2095\n",
      "\ttp: 1400 fp: 2308 tn: 33888 fn: 573\n",
      "threshold: 0.75\n",
      "\tsampled: 2011\n",
      "\ttp: 1172 fp: 1648 tn: 32868 fn: 744\n",
      "threshold: 0.8\n",
      "\tsampled: 2101\n",
      "\ttp: 1026 fp: 1365 tn: 36170 fn: 975\n",
      "threshold: 0.85\n",
      "\tsampled: 2056\n",
      "\ttp: 796 fp: 954 tn: 35765 fn: 1162\n",
      "threshold: 0.9\n",
      "\tsampled: 1952\n",
      "\ttp: 592 fp: 546 tn: 34882 fn: 1270\n",
      "threshold: 0.95\n",
      "\tsampled: 2162\n",
      "\ttp: 439 fp: 354 tn: 38575 fn: 1628\n"
     ]
    }
   ],
   "source": [
    "con_mona_c = sqlite3.connect(mona_comb_db_f)\n",
    "cur_mona_c = con_mona_c.cursor()\n",
    "\n",
    "# randomly sample a subset of the database\n",
    "np.random.seed(469)\n",
    "\n",
    "# test classification accuracy using complete database\n",
    "metrics = []\n",
    "for rep in range(3):\n",
    "    mets = {}\n",
    "    for threshold in np.arange(0.05, 1.0, 0.05):\n",
    "        threshold = round(threshold, 2)\n",
    "        print('threshold:', threshold)\n",
    "        tp, fp, tn, fn = 0, 0, 0, 0\n",
    "        qry_sel1 = 'SELECT name, adduct, pre_mz, msms_mz, msms_i FROM msms;'\n",
    "        qry_sel2 = 'SELECT name, adduct, msms_mz, msms_i FROM msms WHERE pre_mz>=? AND pre_mz<=?;'\n",
    "        i = 0\n",
    "        for name, adduct, pre_mz, msms_mz, msms_i in cur_mona.execute(qry_sel1).fetchall():\n",
    "            if np.random.random() >= 1 / 25:  # sample around 2000 per threshold\n",
    "                # skip this iteration of the loop\n",
    "                continue\n",
    "            mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "            spectrum = spec_norm([np.fromstring(msms_mz.strip('[]'), sep=','), \n",
    "                                  np.fromstring(msms_i.strip('[]'), sep=',')])\n",
    "            for name2, adduct2, msms_mz2, msms_i2 in cur_mona_c.execute(qry_sel2, (mz_min, mz_max)).fetchall():\n",
    "                spectrum2 = [np.frombuffer(msms_mz2), np.frombuffer(msms_i2)]\n",
    "                # label based on similarity threshold\n",
    "                label = spec_entropy_similarity(spectrum, spectrum2) >= threshold\n",
    "                # actual ground truth\n",
    "                known = (name == name2) and (adduct == adduct2)\n",
    "                # count TP, FP, TN, or FN\n",
    "                if label:\n",
    "                    if known:\n",
    "                        tp += 1\n",
    "                    else:\n",
    "                        fp += 1\n",
    "                else:\n",
    "                    if known:\n",
    "                        fn += 1\n",
    "                    else:\n",
    "                        tn += 1\n",
    "            i += 1\n",
    "        print('\\tsampled:', i)\n",
    "        print('\\ttp:', tp, 'fp:', fp, 'tn:', tn, 'fn:', fn)\n",
    "        mets[threshold] = {'sampled': i, 'tp': tp, 'fp': fp, 'tn': tn, 'fn': fn}\n",
    "    metrics.append(mets)\n",
    "    \n",
    "with open('MoNA_combined_spectra_match_test.json', 'w') as jf:\n",
    "    json.dump(metrics, jf, indent=4)\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b25e1883",
   "metadata": {},
   "source": [
    "## Plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "6ba1f7cc",
   "metadata": {},
   "outputs": [],
   "source": [
    "con_nist_c = sqlite3.connect(nist_comb_db_f)\n",
    "cur_nist_c = con_nist_c.cursor()\n",
    "\n",
    "con_mona_c = sqlite3.connect(mona_comb_db_f)\n",
    "cur_mona_c = con_mona_c.cursor()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "4b9f4a8c",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "d2f4a422",
   "metadata": {},
   "outputs": [],
   "source": [
    "# compile the full match results (threshold, FPR, TPR)\n",
    "full_match_results = []\n",
    "for jsonf in ['nist_full_match_test.json', 'nist_full_match_test_2.json']:\n",
    "    with open(jsonf, 'r') as jf:\n",
    "        data = json.load(jf)\n",
    "    for rep in data:\n",
    "        for threshold in rep:\n",
    "            d = rep[threshold]\n",
    "            full_match_results.append([float(threshold), d['fp'] / (d['fp'] + d['tn']), d['tp'] / (d['tp'] + d['fn'])])\n",
    "            \n",
    "fm_thr, fm_fpr, fm_tpr = np.array(sorted(full_match_results, key=lambda x: x[1])).T\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "5e08309e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# compile the exclude target results (threshold, FPR, TPR)\n",
    "exclude_target_results = []\n",
    "for jsonf in ['nist_exclude_target_match_test.json', 'nist_exclude_target_match_test_2.json']:\n",
    "    with open(jsonf, 'r') as jf:\n",
    "        data = json.load(jf)\n",
    "    for rep in data:\n",
    "        for threshold in rep:\n",
    "            d = rep[threshold]\n",
    "            exclude_target_results.append([float(threshold), d['fp'] / (d['fp'] + d['tn']), d['tp'] / (d['tp'] + d['fn'])])\n",
    "            \n",
    "xt_thr, xt_fpr, xt_tpr = np.array(sorted(exclude_target_results, key=lambda x: x[1])).T\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "1837105a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# compile the combined spectra results (threshold, FPR, TPR)\n",
    "combined_spectra_results = []\n",
    "for jsonf in ['nist_combined_spectra_match_test.json']:\n",
    "    with open(jsonf, 'r') as jf:\n",
    "        data = json.load(jf)\n",
    "    for rep in data:\n",
    "        for threshold in rep:\n",
    "            d = rep[threshold]\n",
    "            combined_spectra_results.append([float(threshold), d['fp'] / (d['fp'] + d['tn']), d['tp'] / (d['tp'] + d['fn'])])\n",
    "            \n",
    "cs_thr, cs_fpr, cs_tpr = np.array(sorted(combined_spectra_results, key=lambda x: x[1])).T\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "30504583",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 400x400 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# plot ROC curves\n",
    "fig, ax = plt.subplots(figsize=(4, 4))\n",
    "ax.plot(fm_fpr, fm_tpr, 'g-', lw=1.5, label='FM')\n",
    "ax.plot(xt_fpr, xt_tpr, 'b-', lw=1.5, label='XT')\n",
    "ax.plot(cs_fpr, cs_tpr, 'm-', lw=1.5, label='CS')\n",
    "ax.plot([0, 1], [0, 1], 'k--', lw=0.75, zorder=-1)\n",
    "ax.legend(frameon=False)\n",
    "for d in ['top', 'right']:\n",
    "    ax.spines[d].set_visible(False)\n",
    "ax.set_xlabel('FPR')\n",
    "ax.set_ylabel('TPR')\n",
    "ax.set_xlim([0, 1.05])\n",
    "ax.set_ylim([0, 1.05])\n",
    "plt.show()\n",
    "plt.close()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "id": "fc210b34",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "completed 100 outer loops\n",
      "completed 200 outer loops\n",
      "completed 300 outer loops\n",
      "completed 400 outer loops\n",
      "completed 500 outer loops\n",
      "completed 600 outer loops\n",
      "completed 700 outer loops\n",
      "completed 800 outer loops\n",
      "completed 900 outer loops\n",
      "completed 1000 outer loops\n",
      "# of between comparisons: 73900\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(420)\n",
    "\n",
    "# compute similarities between different combined spectra\n",
    "nist_between_sims = [] \n",
    "qry_sel1 = 'SELECT name, adduct, pre_mz, msms_mz, msms_i FROM msms'\n",
    "qry_sel2 = 'SELECT msms_mz, msms_i FROM msms WHERE NOT (name==? AND adduct==?) AND pre_mz>=? AND pre_mz<=?'\n",
    "i = 0\n",
    "for name, adduct, pre_mz, msms_mz, msms_i in cur_nist_c.execute(qry_sel1).fetchall():\n",
    "    if np.random.random() >= 1 / 37:  # sample around 1000\n",
    "        # skip this iteration of the loop\n",
    "        continue\n",
    "    spectrum = [np.frombuffer(msms_mz), np.frombuffer(msms_i)]\n",
    "    mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "    for msms_mz2, msms_i2 in cur_nist_c.execute(qry_sel2, (name, adduct, mz_min, mz_max)).fetchall():\n",
    "        spectrum2 = [np.frombuffer(msms_mz2), np.frombuffer(msms_i2)]\n",
    "        nist_between_sims.append(spec_entropy_similarity(spectrum, spectrum2))\n",
    "    i += 1\n",
    "    if i % 100 == 0:\n",
    "        print('completed', i, 'outer loops')\n",
    "\n",
    "# fetch mean similarity within combined spectra\n",
    "nist_within_sims = [_[0] for _ in cur_nist_c.execute(\"SELECT mean_spec_sim FROM msms\").fetchall()]\n",
    "\n",
    "print('# of between comparisons:', len(nist_between_sims))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "cbd8cb61",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 200x300 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(figsize=(2, 3))\n",
    "props = lambda c: {k: {'color': c, 'markeredgecolor': c} for k in ['boxprops', 'medianprops', 'flierprops', 'whiskerprops', 'capprops']}\n",
    "w = 0.5\n",
    "ax.boxplot(nist_between_sims, widths=w, positions=[1], showfliers=False, **props('orange'))\n",
    "ax.boxplot(nist_within_sims, widths=w, positions=[2], showfliers=False, **props('purple'))\n",
    "ax.set_ylabel('spectral similarity')\n",
    "ax.set_xticks([1, 2])\n",
    "#ax.set_ylim([0, 1])\n",
    "ax.set_xticklabels(['between', 'within'], rotation=60)\n",
    "for d in ['top', 'right']:\n",
    "    ax.spines[d].set_visible(False)\n",
    "#plt.tight_layout()\n",
    "plt.show()\n",
    "plt.close()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "ed0f0bfb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 400x400 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# compile the full match results (threshold, FPR, TPR)\n",
    "full_match_results = []\n",
    "for jsonf in ['MoNA_full_match_test.json']:\n",
    "    with open(jsonf, 'r') as jf:\n",
    "        data = json.load(jf)\n",
    "    for rep in data:\n",
    "        for threshold in rep:\n",
    "            d = rep[threshold]\n",
    "            full_match_results.append([float(threshold), d['fp'] / (d['fp'] + d['tn']), d['tp'] / (d['tp'] + d['fn'])])\n",
    "            \n",
    "fm_thr, fm_fpr, fm_tpr = np.array(sorted(full_match_results, key=lambda x: x[1])).T\n",
    "\n",
    "\n",
    "# compile the exclude target results (threshold, FPR, TPR)\n",
    "exclude_target_results = []\n",
    "for jsonf in ['MoNA_exclude_target_match_test.json']:\n",
    "    with open(jsonf, 'r') as jf:\n",
    "        data = json.load(jf)\n",
    "    for rep in data:\n",
    "        for threshold in rep:\n",
    "            d = rep[threshold]\n",
    "            exclude_target_results.append([float(threshold), d['fp'] / (d['fp'] + d['tn']), d['tp'] / (d['tp'] + d['fn'])])\n",
    "            \n",
    "xt_thr, xt_fpr, xt_tpr = np.array(sorted(exclude_target_results, key=lambda x: x[1])).T\n",
    "\n",
    "# compile the combined spectra results (threshold, FPR, TPR)\n",
    "combined_spectra_results = []\n",
    "for jsonf in ['MoNA_combined_spectra_match_test.json']:\n",
    "    with open(jsonf, 'r') as jf:\n",
    "        data = json.load(jf)\n",
    "    for rep in data:\n",
    "        for threshold in rep:\n",
    "            d = rep[threshold]\n",
    "            combined_spectra_results.append([float(threshold), d['fp'] / (d['fp'] + d['tn']), d['tp'] / (d['tp'] + d['fn'])])\n",
    "            \n",
    "cs_thr, cs_fpr, cs_tpr = np.array(sorted(combined_spectra_results, key=lambda x: x[1])).T\n",
    "\n",
    "# plot ROC curves\n",
    "fig, ax = plt.subplots(figsize=(4, 4))\n",
    "ax.plot(fm_fpr, fm_tpr, 'g-', lw=1.5, label='FM')\n",
    "ax.plot(xt_fpr, xt_tpr, 'b-', lw=1.5, label='XT')\n",
    "ax.plot(cs_fpr, cs_tpr, 'm-', lw=1.5, label='CS')\n",
    "ax.plot([0, 1], [0, 1], 'k--', lw=0.75, zorder=-1)\n",
    "ax.legend(frameon=False)\n",
    "for d in ['top', 'right']:\n",
    "    ax.spines[d].set_visible(False)\n",
    "ax.set_xlabel('FPR')\n",
    "ax.set_ylabel('TPR')\n",
    "ax.set_xlim([0, 1.05])\n",
    "ax.set_ylim([0, 1.05])\n",
    "plt.show()\n",
    "plt.close()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "id": "e485d8d5",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "completed 100 outer loops\n",
      "completed 200 outer loops\n",
      "completed 300 outer loops\n",
      "completed 400 outer loops\n",
      "completed 500 outer loops\n",
      "completed 600 outer loops\n",
      "completed 700 outer loops\n",
      "completed 800 outer loops\n",
      "completed 900 outer loops\n",
      "completed 1000 outer loops\n",
      "completed 1100 outer loops\n",
      "completed 1200 outer loops\n",
      "completed 1300 outer loops\n",
      "completed 1400 outer loops\n",
      "completed 1500 outer loops\n",
      "completed 1600 outer loops\n",
      "completed 1700 outer loops\n",
      "completed 1800 outer loops\n",
      "completed 1900 outer loops\n",
      "completed 2000 outer loops\n",
      "completed 2100 outer loops\n",
      "completed 2200 outer loops\n",
      "completed 2300 outer loops\n",
      "completed 2400 outer loops\n",
      "completed 2500 outer loops\n",
      "completed 2600 outer loops\n",
      "completed 2700 outer loops\n",
      "completed 2800 outer loops\n",
      "completed 2900 outer loops\n",
      "completed 3000 outer loops\n",
      "completed 3100 outer loops\n",
      "completed 3200 outer loops\n",
      "completed 3300 outer loops\n",
      "completed 3400 outer loops\n",
      "completed 3500 outer loops\n",
      "completed 3600 outer loops\n",
      "completed 3700 outer loops\n",
      "completed 3800 outer loops\n",
      "completed 3900 outer loops\n",
      "completed 4000 outer loops\n",
      "completed 4100 outer loops\n",
      "completed 4200 outer loops\n",
      "completed 4300 outer loops\n",
      "completed 4400 outer loops\n",
      "completed 4500 outer loops\n",
      "completed 4600 outer loops\n",
      "completed 4700 outer loops\n",
      "completed 4800 outer loops\n",
      "completed 4900 outer loops\n",
      "completed 5000 outer loops\n",
      "completed 5100 outer loops\n",
      "completed 5200 outer loops\n",
      "completed 5300 outer loops\n",
      "completed 5400 outer loops\n",
      "completed 5500 outer loops\n",
      "completed 5600 outer loops\n",
      "completed 5700 outer loops\n",
      "completed 5800 outer loops\n",
      "completed 5900 outer loops\n",
      "completed 6000 outer loops\n",
      "completed 6100 outer loops\n",
      "completed 6200 outer loops\n",
      "completed 6300 outer loops\n",
      "completed 6400 outer loops\n",
      "completed 6500 outer loops\n",
      "completed 6600 outer loops\n",
      "completed 6700 outer loops\n",
      "completed 6800 outer loops\n",
      "completed 6900 outer loops\n",
      "completed 7000 outer loops\n",
      "completed 7100 outer loops\n",
      "completed 7200 outer loops\n",
      "completed 7300 outer loops\n",
      "completed 7400 outer loops\n",
      "completed 7500 outer loops\n",
      "completed 7600 outer loops\n",
      "completed 7700 outer loops\n",
      "completed 7800 outer loops\n",
      "completed 7900 outer loops\n",
      "completed 8000 outer loops\n",
      "completed 8100 outer loops\n",
      "completed 8200 outer loops\n",
      "completed 8300 outer loops\n",
      "completed 8400 outer loops\n",
      "completed 8500 outer loops\n",
      "completed 8600 outer loops\n",
      "completed 8700 outer loops\n",
      "completed 8800 outer loops\n",
      "completed 8900 outer loops\n",
      "completed 9000 outer loops\n",
      "completed 9100 outer loops\n",
      "completed 9200 outer loops\n",
      "completed 9300 outer loops\n",
      "completed 9400 outer loops\n",
      "completed 9500 outer loops\n",
      "completed 9600 outer loops\n",
      "completed 9700 outer loops\n",
      "completed 9800 outer loops\n",
      "completed 9900 outer loops\n",
      "completed 10000 outer loops\n",
      "completed 10100 outer loops\n",
      "completed 10200 outer loops\n",
      "completed 10300 outer loops\n",
      "completed 10400 outer loops\n",
      "completed 10500 outer loops\n",
      "completed 10600 outer loops\n",
      "# of between comparisons: 205542\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(420)\n",
    "\n",
    "# compute similarities between different combined spectra\n",
    "mona_between_sims = [] \n",
    "qry_sel1 = 'SELECT name, adduct, pre_mz, msms_mz, msms_i FROM msms'\n",
    "qry_sel2 = 'SELECT msms_mz, msms_i FROM msms WHERE NOT (name==? AND adduct==?) AND pre_mz>=? AND pre_mz<=?'\n",
    "i = 0\n",
    "for name, adduct, pre_mz, msms_mz, msms_i in cur_mona_c.execute(qry_sel1).fetchall():\n",
    "    if np.random.random() >= 1.1:  # sample around 1000\n",
    "        # skip this iteration of the loop\n",
    "        continue\n",
    "    spectrum = [np.frombuffer(msms_mz), np.frombuffer(msms_i)]\n",
    "    mz_min, mz_max = pre_mz - 0.1, pre_mz + 0.1 \n",
    "    for msms_mz2, msms_i2 in cur_mona_c.execute(qry_sel2, (name, adduct, mz_min, mz_max)).fetchall():\n",
    "        spectrum2 = [np.frombuffer(msms_mz2), np.frombuffer(msms_i2)]\n",
    "        mona_between_sims.append(spec_entropy_similarity(spectrum, spectrum2))\n",
    "    i += 1\n",
    "    if i % 100 == 0:\n",
    "        print('completed', i, 'outer loops')\n",
    "\n",
    "# fetch mean similarity within combined spectra\n",
    "mona_within_sims = [_[0] for _ in cur_mona_c.execute(\"SELECT mean_spec_sim FROM msms\").fetchall()]\n",
    "\n",
    "print('# of between comparisons:', len(mona_between_sims))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "id": "088ac81d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPYAAAE/CAYAAAB1gj4AAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy88F64QAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAyDUlEQVR4nO3dfVyN9/8H8NepzilZN5KUobCU5rbc1NyLaO6HGEvG3GyGYV/3Npm5mRnViMxqDbGUu8lN7vkpdxUjFIqklqJTdHvq8/ujb9fXEXWuOqdzztX7+Xj0yLnOp8v72ry6rnPdfN4ixhgDIURQdNRdACFE+SjYhAgQBZsQAaJgEyJAFGxCBIiCTYgAUbAJESAKthIwxpCTkwO6JYBoCgq2EuTm5sLExAS5ubnqLoUQABRsQgSJgk2IAFGwCREgCjYhAkTBJkSAKNiECBAFmxAB0qpgnz9/HkOHDkWTJk0gEolw4MCBSseHh4djwIABaNSoEYyNjeHi4oLjx4/LjQkKCoJIJKrwVVBQoMItIUS1tCrYr169QocOHfDrr78qNP78+fMYMGAAIiIicP36dfTt2xdDhw5FbGys3DhjY2OkpaXJfRkYGKhiEwipFXrqLoAPd3d3uLu7Kzx+06ZNcq9Xr16NgwcP4vDhw+jUqRO3XCQSwdLSUuH1FhYWorCwkHudk5Oj8M8S9SnOK0bm3cwqx8nyZchOzoapjSn06lUdEXN7c4gNxcooUWm0Ktg1VVpaitzcXJiZmcktf/nyJaytrVFSUoKOHTvihx9+kAv+m9asWQNvb29Vl0uULPNuJgKcApS+3mnXp8HK0Urp660JkbZOZigSibB//36MGDFC4Z9Zv3491q5dizt37sDCwgIAEB0djfv376Ndu3bIycmBj48PIiIicOPGDdja2r51PW/bYzdr1gxSqRTGxsY12i6iOorusTPvZCL8s3B8svMTmLcxr3I87bHVKCQkBCtWrMDBgwe5UAOAs7MznJ2dudfdu3eHo6Mj/Pz84Ovr+9Z16evrQ19fX+U1E+USG4p57VnN25hr3J5YUXUi2Hv37sWUKVMQGhqK/v37VzpWR0cHXbp0QWJiYi1VR4jyadVZ8eoICQnBpEmTsHv3bgwePLjK8YwxxMXFwcpKO39TEwJo2R775cuXuH//Pvc6KSkJcXFxMDMzQ/PmzbF48WKkpqYiODgYQFmoJ06cCB8fHzg7OyM9PR0AUK9ePZiYmAAAvL294ezsDFtbW+Tk5MDX1xdxcXHYvHlz7W8gIcrCtMiZM2cYgApfXl5ejDHGvLy8WO/evbnxvXv3rnQ8Y4x98803rHnz5kwikbBGjRoxNzc3dunSJV51SaVSBoBJpVIlbCVRt6fXn7IVWMGeXn+q7lKqTWvPimuSnJwcmJiY0FlxgUiLSUOAU4BGXsZSlOA/YxNSF1GwCREgCjYhAkTBJkSAKNiECBAFmxABomATIkAUbEIEiIJNiABRsAkRIAo2IQJEwSZEgCjYhAgQBZsQAaJgEyJAFGxCBIiCTYgAUbAJESAKNiECpFXB5tttEwDOnTsHJycnGBgYoGXLlti6dWuFMWFhYXBwcIC+vj4cHBywf/9+FVRPSO3RqmDz7baZlJSEjz/+GD179kRsbCyWLFmC2bNnIywsjBsTFRWFsWPHwtPTEzdu3ICnpyc8PDxw+fJlVW0GIaqn7mlSqwsA279/f6VjFixYwOzt7eWWTZ8+nTk7O3OvPTw82KBBg+TGDBw4kI0bN07hWmj6YWERwvTDWtUwgK+oqCi4ubnJLRs4cCB27NiB4uJiiMViREVFYe7cuRXGvNmC93XURlczZSVmoSi3qMbrybyTKfddGSRGEjS0bai09VVF0MFOT09H48aN5ZY1btwYMpkMmZmZsLKyeueY8q4hb0NtdDVP+s10bOuwTanrDP8sXKnrm35jOizbK96HvSa06jN2dYhEIrnX7L/9EV5f/rYxby573eLFiyGVSrmvlJQUJVZMquPZP8/UXUKVarNGQe+xLS0tK+x5MzIyoKenh4YNG1Y65s29+Ouoja7mMbUxBQCFe1rXpvJ+2+U11gZBB9vFxQWHDx+WW3bixAl07twZYrGYGxMZGSn3OfvEiRP46KOParVWUjN69cr+KWtyT+vyGmvl76q1v0kJ+HbbnDFjBn799VfMmzcPU6dORVRUFHbs2IGQkBBuHXPmzEGvXr2wbt06DB8+HAcPHsTJkydx8eLFWt8+QpRG3afl+eDbbZMxxs6ePcs6derEJBIJs7GxYf7+/hXWGxoayuzs7JhYLGb29vYsLCyMV110uUv9NPkSlTpq06o9dp8+fbiTX28TFBRUYVnv3r0RExNT6XpHjx6N0aNH17Q8QjSG4M+KE1IXUbAJESAKNiECRMEmRIAo2IQIEAWbEAGiYBMiQLyvY9vY2GDy5MmYNGkSmjdvroqaiCJkeUDO3arHleQDL5OB92wA3XqVjzW2B/QMlVEdUTPewZ4/fz6CgoKwcuVK9O3bF1OmTMHIkSPpoYjalnMXOOak3HUOug6YOSp3nUQteAd71qxZmDVrFm7cuIHff/8ds2fPxldffYXx48dj8uTJcHSkfxi1wti+LIhVkd4Boj4DXHYCJm2qXicRhGrfUtqhQwf4+Pjg559/xpYtW7Bw4UL4+/ujbdu2mDNnDj7//PNKn2kmNaRnyG/vatKG9sZ1SLWDXVxcjP379yMwMBCRkZFwdnbGlClT8PTpUyxduhQnT57E7t27lVkrIURBvIMdExODwMBAhISEQFdXF56enti4cSPs7f93GOfm5oZevXoptVBCiOJ4B7tLly4YMGAA/P39MWLECG7Cgtc5ODhg3LhxSimQEMIf72A/fPgQ1tbWlY6pX78+AgMDq10UIaRmeN+g0rdvX2RlZVVYnp2djZYtWyqlKEJIzfAOdnJyMkpKSiosLywsRGpqqlKKIoTUjMKH4ocOHeL+fPz4cZiYmHCvS0pKcOrUKdjY2Ci1OEIUVZxXDABIi0mr8bpk+TJkJ2fD1MZUKRMQPrtT+1MjK1z1iBEjAJTNwe3l5SX3nlgsho2NDTZs2KDU4ghRVObdsq4dh6cermKk+kiMJLX2dykc7NLSUgBAixYtcPXqVZiba9bczaRusx9RdrnV3N4cYsOKV2r4KJ8HXJlzlNd2ix+tmqWUMcY2b97MbGxsmL6+PnN0dGTnz59/51gvL6+3zmrq4ODAjQkMDHzrmPz8fIVr0uhZSrOuM7YLZd+JQjR5xlNFKbTH9vX1xbRp02BgYABfX99Kx86ePbv6v2WqsHfvXnzzzTfYsmULunfvjm3btsHd3R3x8fFvfdLMx8cHa9eu5V7LZDJ06NABY8aMkRtnbGyMe/fuyS0zMDBQzUYQUgsUCvbGjRsxYcIEGBgY4JdffnnnPeAikUilwf7ll18wZcoUfPHFFwCATZs24fjx4/D398eaNWsqjDcxMZE7yXfgwAG8ePECn3/+eYW6LS0Vb5ZG3TaJplMo2ElJSdyfk5OTVVVLpYqKinD9+nUsWrRIbrmbmxsuXbqk0Dp27NiB/v37V7jB5uXLl7C2tkZJSQk6duyIH374AZ06dXrneqjbJtF0vK5jFxcXo2XLloiPj1dVPe+UmZmJkpIS3i1vy6WlpeHo0aPc3r6cvb09goKCcOjQIYSEhMDAwADdu3dHYmLiO9dF3TaJpuN1kU4sFqOwsFCtj2PybXlbLigoCKamptxlu3LOzs5wdnbmXnfv3h2Ojo7w8/N75/kE6rZJNB3vO89mzZqFdevWQSaTqaKedzI3N4euri7vlrdAWfh///13eHp6QiKp/Fqijo4OunTpUukemxBNx/u2msuXL+PUqVM4ceIE2rVrh/r168u9Hx4errTiXieRSODk5ITIyEiMHDmSWx4ZGYnhw4dX+rPnzp3D/fv3MWXKlCr/HsYY4uLi0K5duxrXTIi68A62qakpRo0apYpaqjRv3jx4enqic+fOcHFxQUBAAB4/fowZM2YAQIU2uuV27NiBbt26oW3bthXW6e3tDWdnZ9ja2iInJwe+vr6Ii4vD5s2ba2WbCFEF3sFW5+OYY8eORVZWFlauXIm0tDS0bdsWERER3FnutLQ0PH78WO5npFIpwsLC4OPj89Z1ZmdnY9q0aUhPT4eJiQk6deqE8+fPo2vXrirfHkJURcRYJX1piUJycnJgYmICqVQKY2NjdZcj73lM2WymNAOpwtJi0hDgFIBp16fBytFK3eVUS7UeXdm3bx/++usvPH78GEVFRXLvVdWLmhCierzPivv6+uLzzz+HhYUFYmNj0bVrVzRs2BAPHz6Eu7u7KmokhPDEO9hbtmxBQEAAfv31V0gkEixYsACRkZGYPXs2pFKpKmokhPDEO9iPHz/GRx99BACoV68ecnNzAQCenp4ICQlRbnWEkGrhHWxLS0tuzjNra2tER0cDKLufnM7DEaIZeAe7X79+OHy4bJaKKVOmYO7cuRgwYADGjh0rd+MIIUR9eJ8VDwgI4GZTmTFjBszMzHDx4kUMHTqUu1GEEKJevIOto6MDHZ3/7eg9PDzg4eGh1KIIITWjULBv3ryp8Arbt29f7WIIIcqhULA7duwIkUhU5ckxkUj01jnHCSG1i/cMKoQQzadQsKvq1UUI0SwKBfvQoUNwd3eHWCyW6wjyNsOGDVNKYYSQ6lMo2CNGjEB6ejosLCwqTC30OvqMTYhmUCjY5det3/wzIUQz8b7zjBCi+ar1PPaVK1dw9uxZZGRkVNiD//LLL0opjBBSfbyDvXr1aixbtgx2dnZo3Lix3NS/6pyWmBDyP7yD7ePjg99//x2TJk1SQTmEEGXg/RlbR0cH3bt3V0UtCtmyZQtatGgBAwMDODk54cKFC+8ce/bsWYhEogpfd+/elRsXFhYGBwcH6Ovrw8HBAfv371f1ZhCiUryDPXfuXLVNzVvebXPp0qWIjY1Fz5494e7uXmFm0jfdu3cPaWlp3JetrS33XlRUFMaOHQtPT0/cuHEDnp6e8PDwwOXLl1W9OYSoDO9ZSktLSzF48GAkJCTAwcEBYrF8k3FVNQwAgG7dusHR0RH+/v7csjZt2mDEiBFv7bZ59uxZ9O3bFy9evICpqelb1zl27Fjk5OTg6NGj3LJBgwahQYMGCs8IQ7OUCosQZimtVoufM2fOoHXr1mjYsCHXqvbNlrXKVt5t083NTW65It02O3XqBCsrK7i6uuLMmTNy70VFRVVY58CBAytdZ2FhIXJycuS+CNEkvE+eBQcHIywsDIMHD1ZFPe9UnW6bVlZWCAgIgJOTEwoLC/Hnn3/C1dUVZ8+eRa9evQAA6enpvDt4Uhtdoul4B9vMzAytWrVSRS0K4dNt087ODnZ2dtxrFxcXpKSk4Oeff+aCzXedQFkroXnz5nGvc3Jy0KxZM17bQWpfcV4xMu9mVjku806m3PeqmNubQ2wornpgLeId7BUrVuD7779HYGAgDA0NVVHTW9Wk2+brnJ2dsXPnTu61paUl73VSG13tlHk3EwFOAQqPD/9MsfNFmvhZnHewfX198eDBAzRu3Bg2NjYVTp6pqhNITbptvi42NhZWVv/7n+Di4oLIyEjMnTuXW3bixAluimUiHOb25ph2fVqV42T5MmQnZ8PUxhR69aqOiLm9uTLKUyrewa7s6S5V49ttc9OmTbCxscGHH36IoqIi7Ny5E2FhYQgLC+PWOWfOHPTq1Qvr1q3D8OHDcfDgQZw8eRIXL15UyzYS1REbihXeszbrruUfrZiW2bx5M7O2tmYSiYQ5Ojqyc+fOce95eXmx3r17c6/XrVvHWrVqxQwMDFiDBg1Yjx492JEjRyqsMzQ0lNnZ2TGxWMzs7e1ZWFgYr5qkUikDwKRSabW3S2WyrjO2C2XfSZ1B3TaVgK5jE02j0KG4mZkZEhISYG5ujgYNGlR6xvj58+dKK44QUj0KBXvjxo0wMjLi/kxPcRGi2ehQXAnoUJxoGt63lMbExOCff/7hXh88eBAjRozAkiVLUFRUpNTiCCHVwzvY06dPR0JCAgDg4cOHGDt2LAwNDREaGooFCxYovUBCCH+8g52QkICOHTsCAEJDQ9G7d2/s3r0bQUFBcteHCSHqwzvYjDFunrOTJ0/i448/BgA0a9YMmZmK3VtLCFEt3neede7cGatWrUL//v1x7tw57tnopKQkXvdsk0rkJAKyXOWsS3pH/ntN6RkBxrZVjyNqxTvYmzZtwoQJE3DgwAEsXboUH3zwAQBg3759dH+1MuQkAn+3Vv56oz5T3rqGJFC4NZzSLncVFBRAV1e3wkMhdYFSL3eVX55y2QmYtKl5cSX5wMtk4D0bQLdezdYlvVP2C4IunWm8as0r/jYGBgbKWhUBykKtrPA0Ut/kk0Q9qBMIIQJEwSZEgCjYhAgQBZsQAVLo5NnrE/dVhZryEaJ+CgU7NjZWoZXR45yEaAaFgv3mJPuEEM1Gn7EJEaBq3aBy9epVhIaG4vHjxxWewVZl7y5CiGJ477H37NmD7t27Iz4+Hvv370dxcTHi4+Nx+vRplfbuKsenjW54eDgGDBiARo0awdjYGC4uLjh+/LjcmKCgoLe22i0oKFD1phCiMryDvXr1amzcuBF///03JBIJfHx8cOfOHXh4eKB58+aqqJHDt43u+fPnMWDAAEREROD69evo27cvhg4dWuFkoLGxsVyb3bS0NLpFlmg3vvMVGxoasqSkJMYYYw0bNmQ3b95kjDEWHx/PLC0tlTMp8jt07dqVzZgxQ26Zvb09W7RokcLrcHBwYN7e3tzrwMBAZmJiUqO6lDqvuCbPA67JtRE5vPfYZmZmyM0te1b4/fffx61btwAA2dnZyMvLU+ovndfVpI1uudLSUuTm5sLMzExu+cuXL2FtbY2mTZtiyJAhVV7eoza6RNPxDnbPnj0RGRkJAPDw8MCcOXMwdepUfPrpp3B1dVV6geWq00b3TRs2bMCrV6/g4eHBLbO3t0dQUBAOHTqEkJAQGBgYoHv37khMTHznetasWSPXE5w6bRKNw3cXn5WVxVJTUxljjJWUlLB169axoUOHsrlz57Lnz58r/ZCiXGpqKgPALl26JLd81apVzM7Orsqf3717NzM0NGSRkZGVjispKWEdOnRgs2bNeueYgoICJpVKua+UlBQ6FCcahdflLplMhsOHD2PgwIEAAB0dHSxYsKBWZietSRvdvXv3YsqUKQgNDUX//v0rHaujo4MuXbpUusemNrpE0/E6FNfT08OXX36JwsJCVdXzTq+30X1dZGRkpVMyhYSEYNKkSdi9ezcGDx5c5d/DGENcXJxcq11CtA3vG1S6deuG2NhYWFtbq6KeSvFtoxsSEoKJEyfCx8cHzs7O3N6+Xr163DV3b29vODs7w9bWFjk5OfD19UVcXBw2b95c69tHiLLwDvZXX32F+fPn48mTJ3ByckL9+vXl3m/fvr3SinvT2LFjkZWVhZUrVyItLQ1t27ZFREQE90smLS1N7pr2tm3bIJPJMHPmTMycOZNb7uXlhaCgIABlZ/OnTZuG9PR0mJiYoFOnTjh//jy6du2qsu0gRNV4T2aoo1Px6F0kEoExBpFIhJKSEqUVpy1UMpmhJk4YqMm1ETm899hJSUmqqIMQokS8g/3o0SN89NFH0NOT/1GZTIZLly6p5bM3IUQe7xtU+vbt+9bm9lKpFH379lVKUYSQmuG9xy7/LP2mrKysCifSSDWU5Jd9V1ZLHmUqr6m8RqKxFA72J598AqDsRNmkSZPkbtAoKSnBzZs3qcWPMrxMLvuuzJY8yvYymZoQaDiFg11+3ZcxBiMjI9Sr9792MRKJBM7Ozpg6daryK6xr3rMp+66sFj/KVN7ip7xGorEUDnZgYCAAwMbGBv/5z39gaGiosqLqtPL+Wsps8aNsNe0BRlSO98mziRMnIjU1tcLyxMREJCcnK6MmQkgN8Q72pEmT3vr88+XLlzFp0iRl1EQIqSHewY6NjUX37hVPnDg7OyMuLk4ZNRFCaoh3sEUiETeDyuukUmmdvJ2UEE1UrRlU1qxZIxfikpISrFmzBj169FBqcYSQ6uF9g8pPP/2EXr16wc7ODj179gQAXLhwATk5OTh9+rTSCySE8Md7j+3g4ICbN2/Cw8MDGRkZyM3NxcSJE3H37l20bdtWFTUSQniqVieQJk2aYPXq1cquhRCiJNXq3XXhwgV89tln+Oijj7hr2n/++ScuXryo1OIIIdXDO9hhYWEYOHAg6tWrh5iYGG7+s9zcXNqLE6IheB+Kr1q1Clu3bsXEiROxZ88ebvlHH32ElStXKrW4Okn236YLz2OUs76S/LKHNt6zqfmtoJr4xBl5K97BvnfvHnr16lVhubGxMbKzs5VRU92Wc7fs+xUNfqBGz0jdFZAq8A62lZUV7t+/DxsbG7nlFy9eRMuWLZVV1ztt2bIF69evR1paGj788ENs2rSJu+z2NufOncO8efNw+/ZtNGnSBAsWLOBmNS0XFhaG5cuX48GDB2jVqhV+/PFHjBw5UtWb8nZNR5R9N7YH9JTwoE35E1nKelpMzwgwtq35eohq8e0wsG7dOubg4MCio6OZkZERu3DhAtu5cydr1KgR8/PzU247gzfs2bOHicVitn37dhYfH8/mzJnD6tevzx49evTW8Q8fPmSGhoZszpw5LD4+nm3fvp2JxWK2b98+bsylS5eYrq4uW716Nbtz5w5bvXo109PTY9HR0QrXpdSmfMpG3TvqJN7BZoyxJUuWsHr16jGRSMREIhEzMDBgy5YtU3ZtFfDttrlgwQJmb28vt2z69OnM2dmZe+3h4cEGDRokN2bgwIFs3LhxCtdFwSaaplqXu3788UdkZmbiypUriI6OxrNnz/DDDz8o91DiDdXpthkVFVVh/MCBA3Ht2jUUFxdXOqayDp7UbZNoumoFGwAMDQ3RuHFjNGnSBO+9954ya3qr6nTbTE9Pf+t4mUyGzMzMSsdU1sGTum0STcc72DKZDMuXL4eJiQlsbGxgbW0NExMTLFu2jNsLqtKbEymyd0yuWNn4N5fzXefixYshlUq5r5SUFIXrJ6Q28D4r/vXXX2P//v346aef4OLiAqDscHbFihXIzMzE1q1blV4kUL1um5aWlm8dr6enh4YNG1Y6prIOntRtk2g63nvskJAQBAUFYfr06Wjfvj3at2+P6dOn4/fff0dISIgqagRQvW6bLi4uFcafOHECnTt3hlgsrnQMzbhKtBrfs20WFhYsPj6+wvL4+Hhmbm5e89N5lSi/3LVjxw4WHx/PvvnmG1a/fn2WnJzMGGNs0aJFzNPTkxtffrlr7ty5LD4+nu3YsaPC5a7/+7//Y7q6umzt2rXszp07bO3atXS5i2g93sH29vZmn376KSsoKOCWFRQUsAkTJrAVK1Yotbi32bx5M7O2tmYSiYQ5Ojqyc+fOce95eXmx3r17y40/e/Ys69SpE5NIJMzGxob5+/tXWGdoaCizs7NjYrGY2dvbs7CwMF41UbCJpuHdbXPkyJE4deoU9PX10aFDBwDAjRs3UFRUBFdXV7mx4eHhyjqw0GhK7bapbNQhs07iffLM1NQUo0aNkltGl3sI0Sy8g13eOIAQorl4nxXPz89HXl4e9/rRo0fYtGkTTpw4odTCCCHVxzvYw4cPR3BwMAAgOzsbXbt2xYYNGzB8+HD4+/srvUBCCH+8gx0TE8M9Jrlv3z5YWlri0aNHCA4Ohq+vr9ILJITwxzvYeXl5MDIqe9D+xIkT+OSTT6CjowNnZ2c8evRI6QUSQvjjHewPPvgABw4cQEpKCo4fP849GZWRkaF5l3oIqaN4B/u7777Dt99+CxsbG3Tr1o27X/zEiRPo1KmT0gskhPDH+3LX6NGj0aNHD6SlpXE3qACAq6ur+qYTIoTIqVbDAEtLS1haWsot69q1q1IKIoTUXLUnWiCEaC4KNiECRMEmRIAo2IQIEAWbEAGiYBMiQBRsQgSIgk2IAFGwCREgCjYhAqQ1wX7x4gU8PT25tjqenp6V9uMuLi7GwoUL0a5dO9SvXx9NmjTBxIkT8fTpU7lxffr0gUgkkvsaN26cireGENXSmmCPHz8ecXFxOHbsGI4dO4a4uDh4enq+c3xeXh5iYmKwfPlyxMTEIDw8HAkJCRg2bFiFsVOnTkVaWhr3tW3bNlVuCiEqV62HQGrbnTt3cOzYMURHR6Nbt24AgO3bt8PFxQX37t2DnZ1dhZ8xMTGp0OHDz88PXbt2xePHj9G8eXNuuaGhYYWHWgjRZlqxx46KioKJiQkXagBwdnaGiYlJpe1u3ySVSiESiWBqaiq3fNeuXTA3N8eHH36Ib7/9Frm5uZWuh9roEk2nFXvs9PR0WFhYVFhuYWFRabvb1xUUFGDRokUYP3683EwvEyZMQIsWLWBpaYlbt25h8eLFuHHjRoW9/evWrFkDb29v/htCSC1R6x57xYoVFU5cvfl17do1ABVb3QJVt7stV1xcjHHjxqG0tBRbtmyRe2/q1Kno378/2rZti3HjxmHfvn04efIkYmJi3rk+aqNLNJ1a99hff/11lWegbWxscPPmTfz7778V3nv27Fml7W6BslB7eHggKSkJp0+frnJeNkdHR4jFYiQmJsLR8e0tcaiNLtF0ag22ubk5zM3Nqxzn4uICqVSKK1eucDO1XL58GVKptNJ2t+WhTkxMxJkzZ7ie2JW5ffs2iouLYWVlpfiGEKJhtOLkWZs2bTBo0CBMnToV0dHRiI6OxtSpUzFkyBC5M+L29vbYv38/AEAmk2H06NG4du0adu3ahZKSEqSnpyM9PR1FRUUAgAcPHmDlypW4du0akpOTERERgTFjxqBTp07o3r27WraVEKVQc7dPhWVlZbEJEyYwIyMjZmRkxCZMmMBevHghNwYACwwMZIwxlpSUxAC89evMmTOMMcYeP37MevXqxczMzJhEImGtWrVis2fPZllZWbxqoza6RNPwbqNLKqI2ukTTaMWhOCGEHwo2IQJEwSZEgCjYhAgQBZsQAaJgEyJAFGxCBIiCTYgAUbAJESAKNiECRMEmRIAo2IQIEAWbEAGiYBMiQBRsQgSIgk2IAFGwCREgCjYhAqQVDQPIW8jygJy7VY+T3pH/Xhlje0DPsGZ1EY2gNcF+8eIFZs+ejUOHDgEAhg0bBj8/vwrtel43adIk/PHHH3LLunXrhujoaO51YWEhvv32W4SEhCA/Px+urq7YsmULmjZtqpLtUJqcu2VzmSkq6rOqx9C8aIKhNZMZuru748mTJwgICAAATJs2DTY2Njh8+PA7f2bSpEn4999/ERgYyC2TSCQwMzPjXn/55Zc4fPgwgoKC0LBhQ8yfPx/Pnz/H9evXoaurq1BtapnMUNE9dkk+8DIZeM8G0K1X+VjaYwuHeidJVUx8fDwDwKKjo7llUVFRDAC7e/fuO3/Oy8uLDR8+/J3vZ2dnM7FYzPbs2cMtS01NZTo6OuzYsWMK16fR0w+TOkkrTp7VpNvm2bNnYWFhgdatW2Pq1KnIyMjg3rt+/TqKi4vh5ubGLWvSpAnatm1b6Xqp2ybRdFoR7Op223R3d8euXbtw+vRpbNiwAVevXkW/fv1QWFjIrVcikaBBgwZyP9e4ceNK17tmzRqYmJhwX82aNavmlhGiGoLutjl27FgMHjwYbdu2xdChQ3H06FEkJCTgyJEjldZV1Xqp2ybRdILvtvk6KysrWFtbIzExEQBgaWmJoqIivHjxQm6vnZGRUWmzP+q2STSdoLttvikrKwspKSlcJ00nJyeIxWJERkbCw8MDAJCWloZbt27hp59+qsYWEaIh1H32TlGDBg1i7du3Z1FRUSwqKoq1a9eODRkyRG6MnZ0dCw8PZ4wxlpuby+bPn88uXbrEkpKS2JkzZ5iLiwt7//33WU5ODvczM2bMYE2bNmUnT55kMTExrF+/fqxDhw5MJpMpXBudFSeaRmtuUNm1axdmz57NncEeNmwYfv31V7kx9+7dg1QqBQDo6urin3/+QXBwMLKzs2FlZYW+ffti7969MDIy4n5m48aN0NPTg4eHB3eDSlBQkMLXsIGyz+QA6Ow4qTVGRkaVngfSmhtUNNmTJ0/ozDipVVXdDEXBVoLS0lI8ffq0yt+i6pCTk4NmzZohJSVF81r8aiht+G9W1b81rTkU12Q6Ojoaf2+5sbGxxv4j1VTa/N9MK25QIYTwQ8EmRIAo2AKnr6+P77//nm6o4UEI/83o5BkhAkR7bEIEiIJNiABRsAkRIAo2IQJEwSZEgCjYhAgQBZsQAaJgE075LQ10a4NiSktL1V3CO1Gw67jXQyyTyQCUzS/HGKOA/1dJSQmAstlpnz17xs3Dp6Ojo7H/nejOszqspKQEurq6yMzMxMaNG5GUlARjY2N4eXnBxcUFQNleSUeHfv8DwOeff47Lly/j5cuXMDc3x08//YT+/fsD0Lz/TppTCal15bPEeHp64ujRo8jOzkZqaio8PDwwefJkPH36lPvHWld//5dvd3BwMI4ePYoFCxZg/fr1aNOmDdzc3ODu7o6UlBSNCjUA7ZnzjChXSUkJY4yxBw8eMFtbW5aZmckYY+zevXvMz8+Pubi4sFatWrHvv/9ejVWqV2lpKffnbdu2MR8fH+718+fP2cGDB5mLiwvT09NjU6ZMUUeJ70QTLdRR5XuYy5cvo3v37tzn69atW6NFixbo2rUrDh8+DD8/Pzg4OHCzuNYV7LUjlK1bt+LIkSOwtrbmljVo0ABDhw5F586dsXPnTly5ckUdZb4Tfcauw86ePQsPDw8UFRVh//796Nu3r9z72dnZuHXrFnr06KGmCtVPJpNh7ty5CA0NhY6ODg4cOMBNgc3+21iiuLgYpaWlGvWYp4Z9MCC1qWnTpli4cCE+/PBDjBkzBosWLcKLFy+4901NTetcqB88eAB3d3cUFRWhtLQUenp6WLRoEbZv34527dqhT58+mDRpErKysrg5x8RisUaFGqA9NgFw584d7N69G8ePH0dpaSlmzJiBL774Qt1lqcXVq1cRERGB77//Ho8ePYKFhQXq1StrP5yUlISTJ0/C398fT58+xcyZM7F8+XI1V/wO6vyAT2pXeROEwsJClpCQwPbs2cOuXLnC0tPTWWFhIYuIiGAzZsxgrVu3Zq6uriw/P1/NFavXsGHDmEgkYtu2beOWFRQUsLi4OLZixQomFotZYGCg+gqsBO2x66CpU6fi5s2bePr0KYqLizFkyBBs3boVenp6ePbsGcLDw2FoaAhPT091l6pW9+/fx44dO+Dr64sPPvgAPj4+6NOnDwDgxYsXSEhIkGvtrFHU/ZuF1I7yvXVYWBizsLBgUVFRjDHGGjRowNatW8cYY+zff/+t83vpN8lkMnbt2jU2cuRIJhKJ2CeffMJSUlLUXVaV6ORZHVF+M8r27dsxc+ZMODs7w9fXF2ZmZvjqq68AADt37sQff/zBXfqqa8rv/X7+/DkuX76MFy9eoLi4GE5OTti7dy8OHDiAlJQUNG/eHEFBQeottgoU7Dqi/B9tkyZNuEnwvb294e3tjffeew8AkJiYiKtXr0JPr27e3qCjo4Pk5GQMGzYMgwYNQpMmTfD111/jwoULkMlkGDZsGI4fP47169ejZ8+e6i63UvQZu45ZuHAhLl68CFtbW6SlpeH48eMAgOTkZHTs2BGHDx/W+H+0ypSdnQ2pVIqmTZtCV1cXH3/8MXR0dPDdd9/h8ePHWLp0KfLy8jB58mSMHTsWdnZ2vBo2qo26PwsQ1Xv91sj8/Hzm5ubG9PX12eTJk9mrV6/Y3r17mbu7e4W2xHXB4MGDmbW1Ndu3bx9LTExkI0aMYPfu3ZMbs2rVKmZhYcHs7e3ZH3/8oaZK+aFDcQErP/wWiUQoLS1FYmIiDAwMEBgYiJkzZ+LChQto0KABFixYgPfff1/jPzeqwm+//QYnJyeMGTMGK1asQF5eHmJjYwH877bSpUuX4vbt27C1tdWa8w90KF4H+Pj44ODBgygoKMDz58+xe/du2NraIiMjA5mZmTAxMYG9vb26y6x15Y+tAkBUVBRmzZqFmJgYDBw4EGvXroWDgwPEYjF366hWUe8BA1GV8stbf/zxB2vRogX7+eefWWRkJBOJRGzDhg2MMcaKiorUWaJalZaWch9R8vPz2bNnzxhjjAUHB7MGDRqw5s2bs19//ZWlpqbKfZTRFhRsgWvZsiX3uOHatWtZhw4dWH5+PpPJZGzVqlXs8uXLaq5QfcofXf3000/Z3Llz5d6bM2cOE4lErFevXuyvv/5ir169UkeJ1UbBFrAnT56wPn36sOTkZJaXl8dMTU1ZeHg4Y4wxqVTKJk6cyLZs2aLmKtUjKCiIdenShfn7+zORSMTddPL6DTrJycnM1dWViUQi7nl1bVE3L1jWEaampnj+/DmOHj2K6OhoODs7Y+TIkQCAlJQUHDt2DPPnz1dzlbWvuLgYEokELVu2xJw5c9CmTRvuPQMDAxQVFQEArK2tcfLkSSQnJ6Nhw4bqKrd61P2bhahG+efnHTt2sA8//JBJJBL2999/M8YYe/ToEXNzc2OjRo1SZ4lqkZOTw/35r7/+Yqampqxnz57MwcGBLVu2jEmlUu79Y8eOMT8/P3WUWWN0uUtA2H8vcNy7dw/e3t4oKCiAl5cXhg8fDplMhlWrVqFHjx4YNGgQ8vPz8dtvv6m54toVHh4OR0dHbtbRwYMH4/z589iwYQOGDRuGo0ePYsCAAfjtt9/w7NkzTJgwQWvneqPLXQL0+++/Y/78+fjPf/6DJUuWACi7XdTf3x+mpqZo1aoV+vfvj8aNG6u50tp169YtFBQUoHPnzggMDISRkRFGjx4NoOwOtHPnziEiIgIREREoKipCx44duTvztI56DxiIquzdu5cZGRmxCRMmsCdPnqi7HI1SVFTE5s2bxwwNDdnAgQPZ7du3ufdSUlLYhQsX2Pnz51laWpoaq6wZOhQXgNc7UiQkJKCoqAgeHh7Yu3cvnj17hq1btyIjIwMAtObOKVVg/z04DQgIgI2NDf744w8UFhbC2dkZM2fORF5eHpo2bYoePXqgZ8+esLS0VHPF1UfBFoDyGUe/++47DBw4EGPHjoWvry8cHBzg4uKC4OBgbN68GQDq7JNbQNmttbm5uTh79izOnDmDDh064ODBg1i5ciVOnTqF1q1bY9OmTeouUyko2AJRWloKAwMDFBQUoH79+ggJCcGoUaMgk8lgZGSEH374AZ9++il3KaeuMjIywqpVq5Ceno6+ffvi3r17+OabbxAaGorx48dj3rx52LFjh7rLrDE6eSYA7L/3MqelpWH9+vWIjY1FcHAwioqKcP78eWRkZGDdunWYOHGiYPZIfMhkMujp6SEjIwP6+vowMTEBAHz55Ze4ceMGFi5ciOHDhyMrKwu3b99Gr1691FxxzVGwtVh5v6gnT57A0tKSO8xevnw5goKC8Msvv2DMmDEVxtdVnTt3hpGREVxcXODl5YWHDx9i9+7dyMvLw7Jly9CpUyd1l6g0FGwBmDJlCkJCQuDl5YV+/fqhdevWiI+Px7lz5zB69GiucRzTxqeUlOTJkydwdXUFUHaeQV9fHy4uLnjw4AFOnDgBAwMDHD9+XDCTTFCwtVRRUREkEgliY2ORlJSE/Px8bN++HcXFxZDJZHjy5AmKioogEonw999/c90r6irGGM6fP4/ffvsNbm5u6NOnD2JjY5GWloY///wTly5dws2bN9G2bVt1l6oc6rnKRqqj/GmkvLw8xhhjxcXFrEOHDuyff/7hxsTFxbGjR4+yVatWsfbt2zORSMROnz6tlnrVrfxxy9zcXG7ZmTNnWLNmzZinpyd7+fIlt/zNWVO0He2xtVCXLl0wYsQI3L59G0+fPsXZs2eRn5/Pdax4XVxcHDp27Fj7RWqI27dvo127dhg6dCiGDBmCbt264f3338fChQthaGiIb7/9Fs2bN1d3mUpHwdYy5TecBAQE4N9//8XatWsxb948AGUzgpSUlEAikSAvLw+GhoZqrlb9nj9/jnPnzmHPnj1IS0vDq1evkJWVhffffx9PnjzBiBEjsGHDBsFd36dgayHGGPr374+HDx8iNzcXDg4O2LhxI5ycnLgxK1euxODBg+WW1XVJSUlISUlBfHw8QkJCEBcXBysrK9y9e1fdpSkdBVvLsP+e2f7nn39gZmaGuLg4+Pj4IDo6GuPGjYO3tzdOnjyJL774Avn5+XX68lZV4uLiYGZmRofiRDMwxpCfnw+ZTAZjY2Ou39a2bdsQFxeHFi1aYPHixXW2Y2ZV6sL1fAq2liifUfPAgQMIDQ1FZGQkunTpAgcHB/z4448AgIyMDCQkJEAkElVoYk/qFmH/2hKI0tJS6Orq4vnz55gyZQpat26Nffv2IT4+HqmpqZBIJBCLxbCwsEC/fv0o1IT22Nrkyy+/RGpqKg4dOoS0tDS0bt0aZ86cQefOnREREYHS0lK4ublBIpGou1SiZrTH1hJFRUXIz8+Ho6MjAGD06NEYP348OnfuDMYYYmNj8ddff6m5SqIpKNhaQiKR4IMPPkBqairOnTuHpKQkLF++nHv/4MGDcHBwoL01AUCH4lrl5s2bcHV1RVZWFhYuXIg1a9bgyZMnCAkJwfr167lZUggR1u02AsJeexIrPj4eT58+BQDs378fO3fuxE8//YQLFy4gIyMDDRo0QEBAgDrLJRqGgq2hZDIZxGIx1qxZg8DAQGRmZqJhw4Z49uwZJk+ejFOnTuHEiRNo2rQpBgwYAFtbW3WXTDQIHYprmCtXrsDS0hLNmzfHixcvYGFhgeDgYLRr1w6FhYU4cuQI/Pz8MG7cOPj5+am7XKKhKNgaJCsrC66urmjXrh08PDzw6tUrHDx4ELt27eLulHr16hVCQ0Px3XffITw8HJ07d1Zz1UQT0VlxDdKwYUP4+/sjMzMTP//8M06dOoX79+/j1atX3Jj69etj1KhRaNSokfZOZk9UjoKtYVxcXPD3339j+vTpuHbtGq5fv46ZM2fi1q1b3Bh9fX1kZWXB1NRUfYUSjUaH4hrs1atX2LRpE3bv3o1WrVqhY8eOsLGxQUxMDK5evYrLly+ru0SioSjYWuDRo0dYsmQJjhw5gqKiInz66adYtGgRnQkn70SH4lrA2toau3btQmRkJFq2bAkbGxsKNakU7bG1THFxMfLz82FsbKzuUogGo2ATIkB0KE6IAFGwCREgCjYhAkTBJkSAKNiECBAFmxABomATIkAUbEIEiIJNiAD9P8tsCBfDAKH8AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 200x300 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(figsize=(2, 3))\n",
    "props = lambda c: {k: {'color': c, 'markeredgecolor': c} for k in ['boxprops', 'medianprops', 'flierprops', 'whiskerprops', 'capprops']}\n",
    "w = 0.5\n",
    "ax.boxplot(mona_between_sims, widths=w, positions=[1], showfliers=False, **props('orange'))\n",
    "ax.boxplot(mona_within_sims, widths=w, positions=[2], showfliers=False, **props('purple'))\n",
    "ax.set_ylabel('spectral similarity')\n",
    "ax.set_xticks([1, 2])\n",
    "#ax.set_ylim([0, 1])\n",
    "ax.set_xticklabels(['between', 'within'], rotation=60)\n",
    "for d in ['top', 'right']:\n",
    "    ax.spines[d].set_visible(False)\n",
    "#plt.tight_layout()\n",
    "plt.show()\n",
    "plt.close()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cb0a4a72",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}