Untitled

import logging
import requests
import numpy as np
import pandas as pd
import random
import csv
from flask import Flask, request, jsonify, current_app
from scipy.stats import pearsonr

logging.basicConfig(level=logging.INFO)

# Constant Variables
PORT = 5000
HOST = 'localhost'
API_ENDPOINT = 'localhost:5000'
RATING_PATH = './data/ratings.small.csv'
TITLE_PATH = './data/movies.csv'
LINK_PATH = './data/links.csv'

app = Flask(__name__)


# Create rating matrix, list of users, list of movies
def create_rating_matrix(path):
    df = pd.read_csv(path, sep=',', usecols=['userId', 'movieId', 'rating'])
    df = df.pivot(index='userId', columns='movieId', values='rating')
    df = df.fillna(value=0)
    return df.values, df.index.values, df.columns.values


# Import the titles
def create_titles(path):
    df = pd.read_csv(path, sep=',', usecols=['movieId', 'title'])
    return df.values


# Import the links
def create_links(path):
    df = pd.read_csv(path, sep=',', usecols=['movieId', 'imdbId'],
                     dtype={'imdbId': str})
    return df.values


# Create rating matrix, users, movies, titles and links
app.rating_matrix, app.users, app.movies = create_rating_matrix(RATING_PATH)
app.titles = create_titles(TITLE_PATH)
app.links = create_links(LINK_PATH)

# Size down the titles to smaller sample set
temp = []
for title in app.titles:
    if title[0] in app.movies.tolist():
        temp.append(title)
app.titles = np.asarray(temp)

# Size down the links to smaller sample set
temp = []
for link in app.links:
    if link[0] in app.movies.tolist():
        temp.append(link)
app.links = np.asarray(temp)
del temp


# Checks if a User already exists and registers him/her if not
@app.route('/register', methods=['POST'])
def register():
    # Retrieve Data
    data = request.json
    chat_id = data['chat_id']

    # Check if user already exists
    if chat_id in current_app.users.tolist():
        return jsonify({'exists': 1})

    # The chat Id is the new users Id
    current_app.users = np.append(current_app.users, chat_id)

    # Add row of zeros to the rating matrix
    current_app.rating_matrix = np.append(current_app.rating_matrix,
                    np.zeros((1, current_app.rating_matrix.shape[1])),
                                          axis=0)
    return jsonify({'exists': 0})


# Get a random movie that hasn't been rated by the user yet
@app.route('/get_unrated_movie', methods=['POST'])
def get_unrated_movie():
    data = request.json
    chat_id = data['chat_id']
    # Get the ratings of a user by its userId
    user_ratings = current_app.rating_matrix[np.where(current_app.users == chat_id)]
    print(user_ratings)

    # Get a random movie which is not yet rated
    movieId = current_app.movies[random.choice(np.where(user_ratings == 0)[1])]
    # Create the Title
    title = current_app.titles[np.where(current_app.titles == movieId)[0]][0][1]
    # Create URL
    url = 'https://www.imdb.com/title/tt{}/'.format(current_app.links[np.where(current_app.links == movieId)[0]][0][1])
    return jsonify({
        'id': str(movieId),
        'title': title,
        'url': url
    })


# Update the ratingsmatrix after the user has rated it
@app.route('/rate_movie', methods=['POST'])
def rate_movie():
    # Retrieve received data
    data = request.json
    chat_id = data['chat_id']
    movieId = int(data['movieId'])
    rating = int(data['rating'])

    # Updated rating matrix
    current_app.rating_matrix[np.where(current_app.users == chat_id), np.where(current_app.movies == movieId)] = rating

    # Write to csv file, so data isn't lost
    row = [str(chat_id), str(movieId), str(rating), 'N/A']
    with open(RATING_PATH, 'r') as readFile:
        reader = csv.reader(readFile)
        line = list(reader)[-1]
        if line != row:
            with open(RATING_PATH, 'a') as writeFile:
                writer = csv.writer(writeFile)
                writer.writerow(row)
            writeFile.close()
    readFile.close()
    # Return json string
    return jsonify({'status': 'success'})


# Recommend n movies if the user has rated engough movies
@app.route('/recommend', methods=['POST'])
def recommend():
    # Retrieve received data
    data = request.json
    chat_id = data['chat_id']
    top_n = data['top_n']

    # Select the ammount of ratings the user has to submit
    N_RATINGS = 10
    # Check if the User has recommended at least ten movies
    user_idx = np.where(current_app.users == chat_id)
    user_ratings = current_app.rating_matrix[user_idx]
    if np.where(user_ratings != 0)[1].size < N_RATINGS:
        return jsonify({'movies': []})

    # similarity between users function
    def similarity(u1, u2):
        r, _ = pearsonr(current_app.rating_matrix[u1][0], current_app.rating_matrix[u2])
        return r
    # Create list of index and similarity of the 20 users, with the highest similarity in descending order
    similar_neighbours = []
    for i in range(current_app.users.shape[0]):
        if i != user_idx:
            similar_neighbours.append((i, similarity(user_idx, i)))
    similar_neighbours = sorted(similar_neighbours, key=lambda x: x[1], reverse=True)

    # User based collaborative filtering if there are enough recommendations
    # Select the ammount of neighbours to be calculated in the recomendation
    # More neighbours, better prediction, but also more computational time
    N_NEIGHBOURS = 15
    # Create prediction for every single movie with user-based prediction formula
    predictions = []
    # Mean of the ratings of the user
    user_mean = np.mean([r for r in user_ratings[0] if r > 0])

    # Itterate through all the movies
    for movie in current_app.movies:
        prediction = 0
        num = 0
        denom = 0

        # Get the index of the Movie
        movie_idx = np.where(current_app.movies == movie)

        # Check if the movie has already been rated, if yes, then don't put in prediction
        if current_app.rating_matrix[user_idx, movie_idx] == 0:

            # Get all the users which rated the movie
            movie_raters = [neighbour for neighbour in similar_neighbours if
                            current_app.rating_matrix[neighbour[0], movie_idx] > 0][1:N_NEIGHBOURS + 1]

            # Iterate through all the similar neighbours
            for neighbour in movie_raters:
                # index and similarity of the neighbour
                neighbour_idx = neighbour[0]
                neighbour_similarity = neighbour[1]

                # Create the mean of all the ratings of a neighbour
                neighbour_mean = np.mean([r for r in current_app.rating_matrix[neighbour_idx] if r > 0])

                # Get the value the neighbour rated this movie
                neighbour_rating = current_app.rating_matrix[neighbour_idx, movie_idx]

                # Calculate the prediction
                num += neighbour_similarity * (neighbour_rating - neighbour_mean)
                denom += neighbour_similarity
            prediction = user_mean + num / denom
            predictions.append((movie, prediction))

    # Sort the predictions and get the top n
    predictions = sorted(predictions, key=lambda x: x[1], reverse=True)[:top_n]

    # Create response and jsonify it
    movies = []
    for prediction in predictions:
        movies.append({
            'title': current_app.titles[np.where(current_app.titles == prediction[0])[0]][0][1],
            'url': 'https://www.imdb.com/title/tt{}/'.format(
                current_app.links[np.where(current_app.links == prediction[0])[0]][0][1])
        })
    return jsonify({'movies': movies})


# Main function to be run
if __name__ == "__main__":
    app.run(host=HOST, port=PORT)