import tensorflow as tf
import os
import pandas as pd
import numpy as np
import csv
from sklearn.model_selection import train_test_split
import sys
from sacred.observers import MongoObserver
from sacred.observers import FileStorageObserver
from sacred import Experiment

ex = Experiment("444386 sacred_scopes", interactive=True, save_git_info=False)
ex.observers.append(MongoObserver(url='mongodb://admin:IUM_2021@172.17.0.1:27017',db_name='sacred'))
ex.observers.append(FileStorageObserver('training'))
epochs = int(sys.argv[1])

@ex.config
def my_config():
    epoch = epochs
    layerDenseRelu = 256
    layerDropout = 0.01
    layerDenseSoftMax = 1000.0

#ex.add_config("config.json")

@ex.capture
def prepare_data():
    steam=pd.read_csv('data.csv',usecols=[0,1,2,3],names=['userId','game','behavior','hoursPlayed'])
    steam.isnull().values.any()
    steam['userId'] = steam.userId.astype(str)
    purchaseCount = steam[steam["behavior"] != "play"]["game"].value_counts()
    playCount = steam[steam["behavior"] != "purchase"]["game"].value_counts()

    playerPurchaseCount = steam[steam["behavior"] != "play"]["userId"].value_counts()
    playerPlayCount = steam[steam["behavior"] != "purchase"]["userId"].value_counts()

    steam = steam[steam['behavior'] != 'purchase']
    steam = steam.groupby("game").filter(lambda x: len(x)>10)
    size=int(len(steam)/10)

    meanGame = steam[steam["behavior"] != "purchase"].groupby("game").mean()
    meanGame = meanGame.to_dict()
    meanGame = meanGame['hoursPlayed']

    purchaseCount = purchaseCount.to_dict()
    playCount = playCount.to_dict()
    playerPurchaseCount = playerPurchaseCount.to_dict()
    playerPlayCount = playerPlayCount.to_dict()

    steam['meanTime'] = 0;
    steam['purchaseCount'] = 0;
    steam['playCount'] = 0;
    steam['playerPurchaseCount'] =0;
    steam['playerPlayCount'] =0;
    steam['playPercent'] =0;

    for i in steam.index:
        steam.at[i,'meanTime'] = meanGame[steam.at[i,'game']]
        steam.at[i,'purchaseCount'] = purchaseCount[steam.at[i,'game']]
        steam.at[i,'playCount'] = playCount[steam.at[i,'game']]
        steam.at[i,'playerPurchaseCount'] = playerPurchaseCount[steam.at[i,'userId']]
        steam.at[i,'playerPlayCount'] = playerPlayCount[steam.at[i,'userId']]
        steam.at[i,'playPercent'] = playerPlayCount[steam.at[i,'userId']]/playerPurchaseCount[steam.at[i,'userId']]

    steam_train, steam_test = train_test_split(steam, test_size=size, random_state=1, stratify=steam["game"])
    steam_train, steam_dev = train_test_split(steam_train, test_size=size, random_state=1, stratify=steam_train["game"])
    
    games = {}
    for i in steam['game']:
        games[i] = 0

    j=0
    for key,game in games.items():
        games[key]=j
        j=j+1

    for i in steam['game']:
        i = games[i]
        
    invGames = {v: k for k, v in games.items()}

    x_train = steam_train[['hoursPlayed','purchaseCount','playCount','playerPlayCount','playerPurchaseCount']]
    y_train = steam_train['game']

    x_test = steam_test[['hoursPlayed','purchaseCount','playCount','playerPlayCount','playerPurchaseCount']]
    y_test = steam_test['game']


    x_train = np.array(x_train)
    y_train = np.array(y_train)
    x_test = np.array(x_test)
    y_test = np.array(y_test)

    with open('xtest.csv','w',encoding='UTF-8',newline='') as xtest:
        writer = csv.writer(xtest)
        for i in x_test:
            writer.writerow(i)
        
    for i,j in enumerate(y_train):
        y_train[i] = games[j]
        
    for i,j in enumerate(y_test):
        y_test[i] = games[j]
    y_train = np.array(y_train).astype(np.float32)
    y_test = np.array(y_test).astype(np.float32)
    np.savetxt("ytest.csv",y_test,delimiter=",",fmt='%d')
    return x_train, y_train, x_test, y_test, invGames

@ex.main
def my_main(epoch,layerDenseRelu,layerDropout,layerDenseSoftMax,_run):
    x_train, y_train, x_test, y_test, invGames = prepare_data()
    model = tf.keras.models.Sequential([
        tf.keras.layers.Flatten(input_shape=(5,1)),
        tf.keras.layers.Dense(layerDenseRelu, activation='relu'),
        tf.keras.layers.Dropout(layerDropout),
        tf.keras.layers.Dense(layerDenseSoftMax, activation='softmax')
    ])

    model.compile(optimizer='adam',
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])


    model.fit(x_train, y_train, epochs=epoch)
    evaluation = model.evaluate(x_test, y_test)
    _run.log_scalar("training.loss", evaluation[0])
    _run.log_scalar("training.accuracy", evaluation[1])
    
    prediction = model.predict(x_test)
    classes_x=np.argmax(prediction,axis=1)

    rows = []

    for j,i in enumerate(classes_x):
        row = [invGames[i],invGames[y_test[j]]]
        rows.append(row)
    with open('results.csv','w',encoding='UTF-8',newline='') as f:
            writer = csv.writer(f)
            writer.writerow(["predicted", "expected"])
            for row in rows:
                writer.writerow(row)
    
    model.save('./model')
    ex.add_artifact('./model/saved_model.pb')
    

ex.run()