"""Description:- Webapp to run the Hashed key value network on facebook bAbI-task dataset.It takes a joint trained model on entire dataset and populates

a randomly selected story from the dataset to user and also populates the original question from dataset. The question can be changed according to story by

user and depending on question it evaluates the model to generate output answer.It uses flask to run server"""

from __future__ import absolute_import

from __future__ import print_function

from data_utils import load_task, vectorize_data

from sklearn import cross_validation, metrics

from hashed_mem_nw import Hashed_Mem_Nw

from itertools import chain

from six.moves import range, reduce

from hashed_mem_nw import zero_nil_slot, add_gradient_noise

import time

import tensorflow as tf

import numpy as np

import pandas as pd

import random

import glob

import flask

import unicodedata

import warnings

app = flask.Flask(__name__)

timestamp = str(int(time.time()))

data,vocab,word_idx,sentence_size,memory_size,vocab_size=None,None,None,None,None,None#Global variables to store dictionary and data.

#index=1

# Defining initial parameters for the model which can be overwritten from command line.

tf.flags.DEFINE_float("learning_rate", 0.001, "Learning rate for Adam Optimizer.")

tf.flags.DEFINE_float("epsilon", 0.1, "Epsilon value for Adam Optimizer.")

tf.flags.DEFINE_float("l2_lambda", 0.1, "Lambda for l2 loss.")

tf.flags.DEFINE_float("keep_prob", 1.0, "Keep probability for dropout")

tf.flags.DEFINE_float("max_grad_norm", 20.0, "Clip gradients to this norm.")

tf.flags.DEFINE_integer("evaluation_interval", 20, "Evaluate and print results every x epochs")

tf.flags.DEFINE_integer("batch_size", 50, "Batch size for training.")

tf.flags.DEFINE_integer("hops", 3, "Number of hops in the Memory Network.")

tf.flags.DEFINE_integer("epochs", 200, "Number of epochs to train for.")

tf.flags.DEFINE_integer("embedding_size", 40, "Embedding size for embedding matrices.")

tf.flags.DEFINE_integer("memory_size", 50, "Maximum size of memory.")

tf.flags.DEFINE_integer("random_state", None, "Random state.")

tf.flags.DEFINE_string("data_dir", "data/tasks_1-20_v1-2/en/", "Directory containing bAbI tasks")

tf.flags.DEFINE_string("param_output_file", "logs/params_{}.csv".format(timestamp), "Name of output file for model hyperparameters")

tf.flags.DEFINE_string("output_file", "logs/scores_{}.csv".format(timestamp), "Name of output file for final bAbI accuracy scores.")

tf.flags.DEFINE_integer("feature_size", 50, "Feature size")

tf.flags.DEFINE_string("reader", "bow", "Reader for the model")

FLAGS = tf.flags.FLAGS

#Below method processes data and makes a dictionary as well as reverse dictionary.It also initializes initial model.

def getdata():

return data,vocab,word_idx,sentence_size,memory_size,vocab_size

# Fetching a random story and related question from the dataset to be populated.

def story():

return storyS,index,question

# It returns the vector form of story populated to user.

def getVecor(index):

return sV,Q

# Below method vectorize the question sent by the user.

def vectorize_question(question):

return np.array(quest)

# Below method evaluates the story and question on pre trained model to generate output.

def evalModel(sV,Q):

return vocab[test_preds[0]-1]

# Starting the server.

def run():

app.run()

# Rendering the home page to browser.

@app.route('/')

def index():

return flask.render_template("index.html")

# Populating the story when request for new story arrives from browser.

@app.route('/get/story', methods=['GET'])

def get_story():

})

# Passing the predicted answer to user.

@app.route('/get/answer', methods=['GET'])

def get_answer():

})

if __name__ == '__main__':

warnings.filterwarnings('ignore')

data,vocab,word_idx,sentence_size,memory_size,vocab_size=getdata()

run()