basic mix of tensorflow and keras in the same file. Also makes use of tf.data

keras_tensorflow.py

from keras import backend as K
import keras as k
from keras.layers import *
import tensorflow as tf

tf.reset_default_graph()
g=tf.get_default_graph()
sess=tf.Session()
K.set_session(sess)

steps_epoch=100
total_epochs=30
batch_size=4
total_repeats=steps_epoch*total_epochs*batch_size

def train_fn():
    x=tf.constant([
        [[0,1]],
        [[0,0]],
        [[1,0]],
        [[1,1]]
    ],dtype=tf.float32)


    y=tf.constant([
        [[1]],
        [[0]],
        [[1]],
        [[0]]
    ],dtype=tf.int32)
    data=tf.data.Dataset.from_tensor_slices((x,y))
    data=data.repeat(total_repeats)
    data=data.batch(batch_size)
    return data

def val_fn():
    x=tf.constant([
        [[1,1]],
        [[1,0]],
        [[1,0]],
        [[1,1]]
    ],dtype=tf.float32)


    y=tf.constant([
        [[0]],
        [[1]],
        [[1]],
        [[0]]
    ],dtype=tf.int32)
    data=tf.data.Dataset.from_tensor_slices((x,y))
    data=data.repeat(total_repeats)
    data=data.batch(batch_size)
    return data


def mk_iterators(train_fn,val_fn):
    train_data=train_fn()
    val_data=val_fn()
    it=tf.data.Iterator.from_structure(train_data.output_types,train_data.output_shapes)
    val_init=it.make_initializer(val_data)
    train_init=it.make_initializer(train_data)
    x,y=it.get_next()
    return x,y,val_init,train_init

x,y,val_init,train_init=mk_iterators(train_fn,val_fn)


input_=Input(tensor=x)
labs=Input(tensor=y)

net=Dense(24,activation='relu')(input_)
net=Dense(24,activation='relu')(net)
output=Dense(1,activation='sigmoid')(net)
model=k.Model(input_,output)
loss=tf.losses.mean_squared_error(labs,output)
model.add_loss(loss)
model.compile('adam')

sess.run(train_init)
model.fit(steps_per_epoch=steps_epoch,epochs=total_epochs,verbose=0)

# swap datasets and run evaluation step
sess.run(val_init)
model.evaluate(steps=1)


# model.predict() will not work, but you can just call the final layer to get predictions
# (call x and y too if you want to iterate to the next batch for prediction):
sess.run({"input_":x,"OUTPUT":output,"LABS":y})