from keras_tqdm import TQDMNotebookCallback

from keras.models import model_from_json

from keras.preprocessing.image import ImageDataGenerator

from keras.preprocessing import image, sequence

import numpy as np

import pandas as pd

import keras

from keras import backend as K

from keras.utils.data_utils import get_file

from keras.utils import np_utils

from keras.utils.np_utils import to_categorical

from keras.models import Sequential, Model

from keras.layers import Input, Embedding, dot, Reshape, merge, LSTM, Bidirectional

from keras.layers import TimeDistributed, Activation, SimpleRNN, GRU,PReLU

from keras.layers.core import Flatten, Dense, Dropout, Lambda

from keras.regularizers import l2, l1

from keras.layers.normalization import BatchNormalization

from keras.optimizers import SGD, RMSprop, Adam

from keras.metrics import categorical_crossentropy, categorical_accuracy

from keras.layers.convolutional import *

from keras.preprocessing import image, sequence

from keras.preprocessing.text import Tokenizer

from keras.layers import Conv2D, Dense, MaxPool2D, Flatten, Input

from keras.applications import ResNet50, VGG16, VGG19

from sklearn.metrics import log_loss

import time

from sklearn.cross_validation import train_test_split

from sklearn.preprocessing import MinMaxScaler

from keras import regularizers

## #load data

path = "data/"

ratings = pd.read_csv(path+"ratings.csv")   #change this to load ratings.csv file as is appropriate

users = ratings.userId.unique()

movies = ratings.movieId.unique()

userid2idx = {usernum:i for i,usernum in enumerate(users)}

movieid2idx = {movienum:i for i,movienum in enumerate(movies)}


# #convert the dataFrame "ratings" from excel into an array
# #
# #  can also use dataFrame method   .as_matrix()
# #

ratingAra = np.zeros(shape=(len(users),len(movies)))
userIdAra = np.zeros(shape=(len(users),1))
movieIdAra = np.zeros(shape=(len(movies),1))
for i in range(0,len(ratings)):
    userid = ratings[i:i+1].userId.get_values()[0]
    row = userid2idx[userid]
    movie = ratings[i:i+1].movieId.get_values()[0]
    col = movieid2idx[movie]
    userIdAra[row]=userid
    movieIdAra[col]=movie
    ratingAra[row,col]=ratings[i:i+1].rating.get_values()[0]

numUsers = len(users)
numMovies = len(movies)
numUsers, numMovies

### #inner dimension of matrices in product decomposition
numFactors = 50

#######################################
# #model
#

inp_user = Input(shape=(1,), dtype='int64',name='inp_user')
inp_movie = Input(shape=(1,), dtype='int64',name='inp_movie')

u = Embedding(numUsers, numFactors, embeddings_initializer='uniform', embeddings_regularizer=regularizers.l2(1e-4), input_length=1)(inp_user)

m = Embedding(numMovies, numFactors, embeddings_initializer='uniform', embeddings_regularizer=regularizers.l2(1e-4), input_length=1)(inp_movie)

x = dot([u,m], axes=2)  # = dot([u,m], axes=2)  #or = merge([u,m], mode='dot')    #dot product along the second axis of 50 elements

f = Flatten()(x)
model = Model([inp_user, inp_movie],f)

#######################################

model.compile(Adam(0.001), loss='mse')


#
### #set up training & validation sets
#

np.random.seed = 1
mask = np.random.rand(len(ratings)) < 0.8
trainSet= ratings[mask]
valSet = ratings[~mask]


### #model requires numpy arrays
#

userVecT = trainSet.userId.as_matrix()
movieVecT = trainSet.movieId.as_matrix()
ratingVecT = trainSet.rating.as_matrix()
userVecV = valSet.userId.as_matrix()
movieVecV = valSet.movieId.as_matrix()
ratingVecV = valSet.rating.as_matrix()
type(userVecT), userVecT.shape


### #fit model to training set
#

model.fit([userVec, movieVec], ratingVec, batch_size=64, epochs=1,
       validation_data=([userVecV, movieVecV], ratingVecV))

# #  in original code keras 1, this was done:
# #
# #    model.fit([trainSet.userId, trainSet.movieId], trainSet.rating, batch_size=64, epochs=1,
# #            [userIdAra, movieIdAra], ratingAra, batch_size=64, epochs=1,
# #         validation_data=([valSet.userId, valSet.movieId], valSet.rating))

########################################################################
###
### # gives error    InvalidArgumentError: indices[5,0] = 39183 is not in [0, 9066)
###
########################################################################