keras merge

# -*- coding: utf-8 -*-
"""
Created on Sat Nov 12 10:29:43 2016

@author: Administrator
"""
from keras.models import Sequential
from keras.layers import Dense, Activation,Layer,merge
from keras.layers import Merge
from keras.utils.visualize_util import plot
from keras import backend as K
import numpy as np



left_branch = Sequential()
left_branch.add(Dense(32, input_dim=600))

right_branch = Sequential()
right_branch.add(Dense(32, input_dim=600))

def cosine(x):
    axis = len(x[0]._keras_shape)-1
    dot = lambda a, b: K.batch_dot(a, b, axes=axis)
    return dot(x[0], x[1]) / K.sqrt(dot(x[0], x[0]) * dot(x[1], x[1]))
    
def similarity(x):
    return (x[0] * x[1]).sum() / (np.sqrt((x[0] * x[0]).sum()) * np.sqrt((x[1] * x[1]).sum()))*x[0]
    
merged = Merge([left_branch, right_branch], mode='cos',output_shape=lambda x: x[0])

model = Sequential()
model.add(merged)
model.add(Dense(10, activation='softmax'))

model.compile(optimizer='rmsprop',
              loss='categorical_crossentropy',
              metrics=['accuracy'])


from keras.utils.np_utils import to_categorical
data_1 = np.random.random((1000, 600))
data_2 = np.random.random((1000, 600))

# these are integers between 0 and 9
labels = np.random.randint(10, size=(1000, 1))
# we convert the labels to a binary matrix of size (1000, 10)
# for use with categorical_crossentropy
labels = to_categorical(labels, 10)
model.fit([data_1,data_2], labels, nb_epoch=10, batch_size=32)

plot(model, to_file='model2.png',show_shapes=True)