model for age estimation

# In[1]:


import tensorflow as tf
import numpy as np
import pandas as pd
import scipy.io as sio


# In[10]:


data = sio.loadmat('awa_test')
data_seen = data['X_tr']
data_unseen = data['X_te']
seen_semantic = data['S_tr']
latent_dim = 100
learning_rate = 0.001


# In[11]:


X1 = tf.placeholder(tf.float32,shape=[None,1024])
X2 = tf.placeholder(tf.float32,shape=[None,1024])
W = tf.Variable(tf.random_normal([1024,latent_dim]))
b = tf.Variable(tf.random_normal([1,latent_dim]))
X1_proj = tf.cast(tf.add(tf.matmul(X1,W),b),tf.float32)
X2_proj = tf.cast(tf.add(tf.matmul(X2,W),b),tf.float32)


# In[12]:


#For the seen visual, calculate the covariance matrix
temp = tf.matmul(tf.ones([tf.shape(X1_proj)[0],tf.shape(X1_proj)[0]]),X1_proj)*tf.cast(1/tf.shape(X1_proj)[0],tf.float32)
cov1 = tf.subtract(X1_proj,temp)
#For unseen visual, calculate the covariance matrix
temp = tf.matmul(tf.ones([tf.shape(X2_proj)[0],tf.shape(X2_proj)[0]]),X2_proj)*tf.cast(1/tf.shape(X2_proj)[0],tf.float32)
cov2 = tf.subtract(X2_proj,temp)


# In[13]:


loss = tf.norm(tf.subtract(tf.log(cov1),tf.log(cov2)))


# In[20]:


def get_batches(batch_size, dataset, name):
        total_batch = int(dataset.shape[0]/batch_size)
        batches = []
        if total_batch == 0:
            return [dataset]
        for i in range(total_batch):
            if i == total_batch-1:
                batches.append(dataset[i*batch_size:, :])
            else:
                batches.append(dataset[i*batch_size:(i+1)*batch_size, :])
        return batches


# In[ ]:


optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss)
init = tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init)
    for epoch in range(500):
        avg_cost = 0
        total_batch=60
        batches_data_seen = get_batches(100, data_seen, 'train')
        batches_data_unseen = get_batches(100, data_unseen, 'train')
        for i in range(total_batch):
#             batch_x1, _ = tf.train.batch(data_seen,batch_size=100)
#             batch_x2, _ = tf.train.batch(data_unseen,batch_size=100)
			batch_x1 = batches_data_seen[i]
			batch_x2 = batches_data_unseen[i]
			_ , c = sess.run([optimizer, loss], feed_dict = {X1: batch_x1, X2: batch_x2})
			avg_cost +=c
			print ("Epoch:", (epoch+1), "cost =", "{:.5f}".format(avg_cost))
    
	print ("Training complete!")