Feedforward neural network template

template.py

import tensorflow as tf 
import numpy as np 

def get_batch(X, size):
     
     # will return batches ( of size approx 10 ) of tuples ( or list ) of input features 
     # ( i.e. our data collected from the portals ) and corresponding output labels


def getdata(self):
	
	# will fetch data from our aggregated dataset ( combined data from 3 portals in the form of [[input],[corresponding_output]] )



class FeedForward:
	
	def __init__(self,epoch, input_dim, hidden_dim, output_dim, batch_size,learning_rate,window):
		
		self.epoch = epoch
		self.input_dim = input_dim
		self.hidden_dim = hidden_dim
		self.output_dim = output_dim
		self.batch_size = batch_size
		self.learning_rate = learning_rate

		with tf.name_scope('input_to_hidden'):
			
			self.W1 = tf.Variable(tf.random_normal([self.input_dim, self.hidden_dim]),name='W1') 
			self.d = tf.Variable(tf.zeros([self.hidden_dim,1)) # Hidden layer biases
			

		with tf.name_scope('hidden_to_output'):		
			 
			self.W2 = tf.Variable(tf.random_normal([self.hidden_dim, self.output_dim],dtype=float32),name='U') 
			
	
	def main():
	
		# Forward Phase
		Hidden = tf.tanh(tf.matmul(self.W1, input_set) + self.d )
		predicted_output = tf.nn.relu( tf.matmul(Hidden, self.W2) )



		# Backward Phase					       
		loss = -( predicted_output - actual_output ) ** 2  # Squared error loss function
		optimizer = tf.train.GradientDescentOptimizer(self.learning_rate)
		train_op = optimizer.minimize(loss)
		
		X = tf.placeholder(tf.float32, shape=[None, self.input_dim])
		Y = tf.placeholder(tf.float32, shape=[None, self.output_dim])

	
	
		sess.run(tf.global_variables_initializer())			       
		 with tf.Session() as sess:
			for epoch in range(self.epoch):
				X, Y = get_batch()
				sess.run(train_op, feed_dict={ input_set : X, actual_output : Y }