fm.py

import numpy as np
from sklearn.base import BaseEstimator
from keras.layers import Input, Embedding, Dense,Flatten ,Activation, Add, Dot
from keras.models import Model
from keras.regularizers import l2 as l2_reg
from keras import initializers
import itertools

def build_model(max_features,K=8,solver='adam',l2=0.0,l2_fm = 0.0):

    inputs = []
    flatten_layers=[]
    columns = range(len(max_features))
    for c in columns:
        inputs_c = Input(shape=(1,), dtype='int32',name = 'input_%s'%c)
        num_c = max_features[c]
        
        embed_c = Embedding(num_c,K,input_length=1,name = 'embed_%s'%c,embeddings_regularizer=l2_reg(l2_fm))(inputs_c)

        flatten_c = Flatten()(embed_c)

        inputs.append(inputs_c)
        flatten_layers.append(flatten_c)

    fm_layers = []
    for emb1,emb2 in itertools.combinations(flatten_layers, 2):
        dot_layer = Dot(1)([emb1,emb2])
        fm_layers.append(dot_layer)

    for c in columns:
        num_c = max_features[c]
        embed_c = Embedding(num_c,1,input_length=1,name = 'bias_%s'%c,embeddings_regularizer=l2_reg(l2))(inputs[c])

        flatten_c = Flatten()(embed_c)

        fm_layers.append(flatten_c)
        
        
    #flatten = merge(fm_layers,mode='sum')
    flatten = Add()(fm_layers)
    outputs = Activation('sigmoid',name='outputs')(flatten)
    
    model = Model(input=inputs, output=outputs)

    model.compile(
                optimizer=solver,
                loss= 'binary_crossentropy'
              )

    return model
    
    
if __name__ == '__main__':
    from keras.utils import plot_model
    num_users=100
    num_items=100
    k=10
    mdl = build_model([num_users, num_items], K=k)
    mdl.summary()
    
    #plot_model(mdl, to_file="fm.png")
    
    
    print(mdl.predict([[1,2],[1,2]]))