animerec.py

#!/usr/bin/env python3

import sys
import random
from random import randint
from math import sqrt
import rankings
import json
import animelist
import numpy
import types

"""
Main executable. See README.md for details.
"""

if len(sys.argv) < 2 or len(sys.argv) > 3:
    print("usage: animerec username [animeID]")
    exit(0)

predictID = sys.argv[2] if len(sys.argv) == 3 else None

def dot(v1, v2):
    """Dot product implementation"""
    s = 0
    for key in v1:
        if key in v2:
            s += v1[key] * v2[key]
    return s

def normalize(vec):
    """Normalize a vector by computing z-scores"""
    if vec == {}:
        return
    ratings = list()
    for animeID in vec:
        ratings.append(vec[animeID])
    arr = numpy.asarray(ratings)
    mean = numpy.mean(arr)
    sdev = numpy.std(arr)
    for animeID in vec:
        if sdev == 0:
            vec[animeID] = 0
        else:
            vec[animeID] = (vec[animeID] - mean) / sdev
    return (mean, sdev)

x0 = animelist.userToVec(sys.argv[1])
if x0 == {}:
    print("Cannot find " + sys.argv[1] + " on MyAnimeList")
    exit(0)

x = dict()
for animeID in x0:
    x[str(animeID)] = x0[animeID]

if predictID is not None:
    if predictID in x:
        print("You've given it a rating of " + str(x[predictID]) + ". Now let's pretend you haven't watched it...")
        del x[predictID]

mean, sdev = normalize(x)

with open("vecs.json") as f:
    users = json.load(f)

for name in users:
    normalize(users[name])

# Retrieve 30 closest neighbors
similarUsers = rankings.rankings(30)

for name in users:
    if name == sys.argv[1]:
        continue
    v = users[name]
    similarUsers.insert(v, dot(v, x))

scores = dict()
for vec in similarUsers.getList():
    for animeID in vec:
        if animeID in scores:
            scores[animeID].value += vec[animeID]
            scores[animeID].count += 1
        else:
            scores[animeID] = types.SimpleNamespace(value=vec[animeID], count=1)

critical_popularity = 10

if predictID is not None:
    if predictID not in scores:
        print("No data available.")
        exit(0)
    score = scores[predictID]
    if score.count < critical_popularity:
        print("Unpopular series; take prediction with a grain of salt.")
    print("Predicted rating: " + str(round((mean + sdev * score.value / score.count),2)))
else:
    # store up to 40 potential recommendations
    bestAnimes = rankings.rankings(40)
    for animeID in scores:
        score = scores[animeID]
        if score.count >= critical_popularity:
            bestAnimes.insert(animeID, score.value / score.count)

    outputted = 0
    for animeID in bestAnimes.getList():
        if outputted >= 5:
            break
        if animeID in x:
            continue
        print("https://myanimelist.net/anime/" + animeID)
        print(animelist.getTitle(animeID))
        outputted += 1