models.py 文件源码

def _generate_model(self, lembedding, num_classes=2, num_features=128, train_vectors=True):

        model = Sequential()
        if lembedding.vector_box.W is None:
            emb = Embedding(lembedding.vector_box.size,
                            lembedding.vector_box.vector_dim,
                            W_constraint=None,
                            input_length=lembedding.size)
        else:
            emb = Embedding(lembedding.vector_box.size,
                            lembedding.vector_box.vector_dim,
                            weights=[lembedding.vector_box.W], W_constraint=None,
                            input_length=lembedding.size)
        emb.trainable = train_vectors
        model.add(emb)

        model.add(Convolution1D(num_features, 3, init='uniform'))
        model.add(Activation('relu'))
        model.add(MaxPooling1D(2))
        model.add(Dropout(0.25))

        model.add(Convolution1D(num_features, 3, init='uniform'))
        model.add(Activation('relu'))
        model.add(MaxPooling1D(2))
        model.add(Dropout(0.25))

        model.add(Flatten())

        if num_classes == 2:
            model.add(Dense(1, activation='sigmoid'))
            if self.optimizer is None:
                self.optimizer = 'rmsprop'
            model.compile(loss='binary_crossentropy', optimizer=self.optimizer, metrics=["accuracy"])
        else:
            if self.optimizer is None:
                self.optimizer = 'adam'
            model.add(Dense(num_classes, activation='softmax'))
            model.compile(loss='categorical_crossentropy', optimizer=self.optimizer, metrics=["accuracy"])

        return model