02_Lets_Find_A_Model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finding a Machine Learning Model\n",
    "=================\n",
    "\n",
    "There are many different types of models we can use for classification,\n",
    "each of these models have types of problems they are well suited to.\n",
    "The goal of this notebook is to identify algorithms \n",
    "that will effectively classify our dataset\n",
    "which we can then investigate further."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Collating the models\n",
    "\n",
    "The first step here is creating a list of models we would like to test.\n",
    "An excellent property of scikit-learn is \n",
    "that all the algorithms have the same API,\n",
    "allowing us to treat them all in the same way. \n",
    "\n",
    "This is not an exhastive list of all the possible classifiers in scikit-learn,\n",
    "just a smattering for comparison.\n",
    "For a more exhastive list check out [the scikit-learn documentation](http://scikit-learn.org/stable/supervised_learning.html#supervised-learning),\n",
    "and feel free to add more to the list."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LogisticRegression, SGDClassifier\n",
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.naive_bayes import GaussianNB\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.neural_network import MLPClassifier\n",
    "\n",
    "ml_models = {\n",
    "    'LR': LogisticRegression(),\n",
    "    'SGD': SGDClassifier(tol=1e-3, max_iter=1000),\n",
    "    'LDA': LinearDiscriminantAnalysis(),\n",
    "    'DT': DecisionTreeClassifier(),\n",
    "    'KNN': KNeighborsClassifier(),\n",
    "    'NB': GaussianNB(),\n",
    "    'SVM': SVC(),\n",
    "    'NN': MLPClassifier()\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loading the training data\n",
    "\n",
    "We need to load in the training dataset we created in the first notebook.\n",
    "At this point we are interested in two sets of data,\n",
    "\n",
    "- $X$, the input data which is the orientation of the six nearest neighbours\n",
    "- $Y$, the true labelled classification of the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas\n",
    "\n",
    "training_dataset = pandas.read_hdf('data/training_data.h5', key='trimer')\n",
    "X = training_dataset[['orient0', 'orient1', 'orient2', 'orient3', 'orient4', 'orient5']].values\n",
    "Y = training_dataset['class'].values\n",
    "classes = training_dataset['class'].unique()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Testing the Models\n",
    "\n",
    "With a collection of models to test,\n",
    "we now need some method of testing the models to compare them.\n",
    "To perform the initial screening of datasets\n",
    "we are going to break our training data into two groups,\n",
    "\n",
    "- the training set, comprising 80% of the molecules\n",
    "- the validation set, comprising the remaining 20%.\n",
    "\n",
    "This division of the dataset gives us a set of data \n",
    "previously unseen by the algorithms,\n",
    "giving us a method of testing whether\n",
    "the algorithm is acutally learning the underlying features,\n",
    "or just 'memorising' the training data.\n",
    "This division of data will be through a random selection\n",
    "so as not to bias the selection of data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn import model_selection\n",
    "\n",
    "validation_size = 0.20\n",
    "seed = 7\n",
    "\n",
    "selected = model_selection.train_test_split(X, Y, test_size=validation_size, random_state=seed)\n",
    "X_train, X_validation, Y_train, Y_validation = selected"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To get an idea of the models which\n",
    "warrant a further investigation,\n",
    "we can iterate through each of our models.\n",
    "Each model is scored by breaking the training data into `n_splits`,\n",
    "using one of these splits for testing and\n",
    "the remaining splits for training.\n",
    "This process is referred to as *cross validation*\n",
    "and typically the number of splits is 10.\n",
    "For the purposes of this running in a reasonable amount of time,\n",
    "`n_splits` is set to 2."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR   : 0.91845 ± 0.00067\n",
      "SGD  : 0.91503 ± 0.00062\n",
      "LDA  : 0.90854 ± 0.00134\n",
      "DT   : 0.93097 ± 0.00081\n",
      "KNN  : 0.95798 ± 0.00025\n",
      "NB   : 0.93312 ± 0.00084\n",
      "SVM  : 0.95165 ± 0.00066\n",
      "NN   : 0.95561 ± 0.00122\n"
     ]
    }
   ],
   "source": [
    "scoring='accuracy'\n",
    "n_splits = 2\n",
    "# Typically n_splits would be 10 but it runs much slower\n",
    "#n_splits = 10\n",
    "\n",
    "# Iterate through each model in our dictionary of models\n",
    "for name, model in ml_models.items():\n",
    "    kfold = model_selection.KFold(n_splits=n_splits, random_state=seed)\n",
    "    cv_results = model_selection.cross_val_score(model, X_train, Y_train, cv=kfold, scoring=scoring)\n",
    "    print(f'{name:5s}: {cv_results.mean():.5f} ± {cv_results.std():.5f}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Out of all the algorithms tested, \n",
    "there are three that stand out \n",
    "\n",
    "- K-Nearest Neighbours (KNN),\n",
    "- Support Vector Machine (SVM), and\n",
    "- Neural Network (NN)\n",
    "\n",
    "with accuracies in excess of 95%. \n",
    "\n",
    "So with these three algorithms it is likely worth\n",
    "tweaking the algorithms slightly from \n",
    "the defualt paramters in an effort to improve performance.\n",
    "It is also worth understanding which classes\n",
    "each of these algorithms is strongest at classifying.\n",
    "For this additional data we are going to be using a [confusion matrix](http://scikit-learn.org/stable/modules/generated/sklearn.metrics.confusion_matrix.html).\n",
    "In a confusion matrix, \n",
    "the diagonal elements represent the correct classifications,\n",
    "while the off diagonal elements are the values \n",
    "which were incorrectly classified.\n",
    "\n",
    "Below we have a handy function from the scikit-learn documentation\n",
    "that will nicely plot the confusion matrix as a heatmap. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "import itertools\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "%matplotlib inline\n",
    "\n",
    "def plot_confusion_matrix(cm, classes,\n",
    "                          normalize=False,\n",
    "                          title='Confusion matrix',\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    This function prints and plots the confusion matrix.\n",
    "    Normalization can be applied by setting `normalize=True`.\n",
    "    \"\"\"\n",
    "    if normalize:\n",
    "        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "        print(\"Normalized confusion matrix\")\n",
    "    else:\n",
    "        print('Confusion matrix, without normalization')\n",
    "\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    tick_marks = np.arange(len(classes))\n",
    "    plt.xticks(tick_marks, classes, rotation=45)\n",
    "    plt.yticks(tick_marks, classes)\n",
    "\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
    "        plt.text(j, i, format(cm[i, j], fmt),\n",
    "                 horizontalalignment=\"center\",\n",
    "                 color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It is this point where our breaking the data into\n",
    "training and validation sets becomes useful.\n",
    "We can train a classifier using the training set,\n",
    "then generate predictions of the validation dataset\n",
    "using the known values of the validation data\n",
    "to generate the confusion matrix."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Normalized confusion matrix\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1117f5be0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "knn = ml_models['KNN']\n",
    "knn.fit(X_train, Y_train)\n",
    "predictions = knn.predict(X_validation)\n",
    "plot_confusion_matrix(confusion_matrix(Y_validation, predictions, labels=classes), classes, normalize=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Normalized confusion matrix\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x121b0a940>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "svm = ml_models['SVM']\n",
    "svm.fit(X_train, Y_train)\n",
    "predictions = svm.predict(X_validation)\n",
    "plot_confusion_matrix(confusion_matrix(Y_validation, predictions, labels=classes), classes, normalize=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Normalized confusion matrix\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11de362b0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "nn = ml_models['NN']\n",
    "nn.fit(X_train, Y_train)\n",
    "predictions = nn.predict(X_validation)\n",
    "plot_confusion_matrix(confusion_matrix(Y_validation, predictions, labels=classes), classes, normalize=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It is interesting to note that all of the models\n",
    "have the most difficulty with the liquid/crystal characterisation,\n",
    "with the largest proportion of false positives being \n",
    "crystal incorrectly classified as liquid. \n",
    "To make this model we have created persistent\n",
    "it needs to be saved which is done using `joblib`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['data/knn-model.pkl']"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.externals import joblib\n",
    "joblib.dump(ml_models['KNN'], 'data/knn-model.pkl') "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we have a model, we can look at using it in [production](03_ML_in_Production.ipynb) to measure useful quantities."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "MLCrystals",
   "language": "python",
   "name": "mlcrystals"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}