Calculate the outer product/bilinear projection in Keras

outer_product_keras.py

from keras.layers import Lambda
from keras import backend as K
from numpy import newaxis
from keras.models import Model, Input


def outer_product(inputs):
    """
    inputs: list of two tensors (of equal dimensions, 
        for which you need to compute the outer product
    """
    x, y = inputs
    batchSize = K.shape(x)[0]
    outerProduct = x[:,:, newaxis] * y[:,newaxis,:]
    outerProduct = K.reshape(outerProduct, (batchSize, -1))
    # returns a flattened batch-wise set of tensors
    return outerProduct

def main(input_dim=32):
    inputX = Input(shape=(input_dim, ))
    inputY = Input(shape=(input_dim, ))
    bilinearProduct = Lambda(outer_product, output_shape=(input_dim**2, ))([inputX, inputY])
    #tensor 'bilinearProduct' contains the batchwise outer product    
    model = Model(input=[inputX, inputY], output=bilinearProduct)
    #you can run model.predict to get the output of the outer product of each sample in your batch
    return model

coud you show how to call this mode:
`x=np.random.rand(10,1,32)

input_a = np.reshape(x, (10,1, 32))
input_b = np.reshape(x, (10,1,32))
print input_a.shape
print input_b.shape

model=main()

y=model([input_a,input_b])
print(y.shape)`

input_a = np.random.rand(10,32)

input_b = np.random.rand(10,32)

model = main()

model.summary()

y = model.predict([input_a,input_b])

# y will be your output flattened tensor

print(y.shape)

Hope this helps!

what does your newaxis mean?

numpy.newaxis is used to extend the dimension of the vector.

https://stackoverflow.com/questions/29241056/how-does-numpy-newaxis-work-and-when-to-use-it

Excellent.