keras_tf_example.py

import os
os.environ["CUDA_VISIBLE_DEVICES"]="1"

import tensorflow as tf
sess = tf.Session()

from keras import backend as K
K.set_session(sess)

img = tf.placeholder(tf.float32,shape=(None,784))

from keras.layers import Dense
from keras.objectives import categorical_crossentropy
from tensorflow.examples.tutorials.mnist import input_data

# x = Dense(128,activation='relu')(img)
# x = Dense(128,activation='relu')(x)
# preds = Dense(10,activation='softmax')(x)
dense1 = tf.layers.dense(inputs=img,units=128,activation=tf.nn.relu)
dense2 = tf.layers.dense(inputs=dense1,units=128,activation=tf.nn.relu)
preds = tf.layers.dense(inputs=dense2,units=10,activation=tf.nn.softmax)

labels = tf.placeholder(tf.float32,shape=(None,10))
loss = tf.reduce_mean(categorical_crossentropy(labels,preds))

mnist_data = input_data.read_data_sets('MNIST_data',one_hot = True) # 
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(loss)

#builds the graph.
sess.run(tf.global_variables_initializer())

import time
t0 = time.time()
with sess.as_default():
    for i in range(10000):
        batch = mnist_data.train.next_batch(50)
        train_step.run(feed_dict={img:batch[0],labels:batch[1]})
print time.time() - t0 #16.266, native tf: 15.91

### Try with pre-loading the data onto the card
import numpy as np

load_size = 50
num_loads = 10000

all_X = np.zeros((num_loads*load_size,784)) #image
all_y = np.zeros((num_loads*load_size,10))  #classes

for i in range(num_loads):
    batch = mnist_data.train.next_batch(load_size)
    all_X[i*load_size:(i+1)*load_size,:] = batch[0]
    all_y[i*load_size:(i+1)*load_size,:] = batch[1]
    
tf_data_X = tf.constant(all_X,dtype=tf.float32)
tf_data_y = tf.constant(all_y,dtype=tf.float32)
ix = tf.placeholder(shape=(),dtype=tf.int32)

batch_X = tf.slice(tf_data_X,[load_size*ix,0],[load_size,-1])
batch_y = tf.slice(tf_data_y,[load_size*ix,0],[load_size,-1])


## Keras Helpers
# x = Dense(128,activation='relu')(batch_X)
# x = Dense(128,activation='relu')(x)
# preds = Dense(10,activation='softmax')(x)
## pure TF
dense1 = tf.layers.dense(inputs=batch_X,units=128,activation=tf.nn.relu) 
dense2 = tf.layers.dense(inputs=dense1,units=128,activation=tf.nn.relu)
preds = tf.layers.dense(inputs=dense2,units=10,activation=tf.nn.softmax)

loss = tf.reduce_mean(categorical_crossentropy(batch_y,preds))
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(loss)

t0 = time.time()
with sess.as_default():
    sess.run(tf.global_variables_initializer())
    for i in range(1000):
        b_x,b_y,step_update = sess.run([batch_X,batch_y,train_step],feed_dict={ix:i})
print time.time() - t0 # 5.821, native tf: 5.909