from keras.models import Sequential

from keras.layers import Dense

from keras.optimizers import Adam

model = Sequential()

model.add(Dense(50, input_dim=len(train_x[0]),

activation='relu'))

model.add(Dense(8, activation='relu'))

model.add(Dense(len(train_y[0]),

activation='softmax'))

model.summary()

The training of the model is even simpler, as per denition the

intents/utterances dataset used per input is very limited, and there’s no

space at all for creating validation and testing sets.

It basically trains the entire dataset with a sucient number of epochs

to obtain maximum possible accuracy.

model.compile(loss='categorical_crossentropy',

optimizer='adam', metrics=['accuracy'])

model.fit(np.array(train_x), np.array(train_y),

epochs=400)

Once the Keras/TensorFlow model is trained, this is easily exported to

Core ML using the Apple CoreMLTools python library:

import coremltools

coreml_model =

coremltools.converters.keras.convert(model,

input_names="embeddings", output_names="entities")

coreml_model.save('SwiftNLC.mlmodel')

Core ML Swift Wrapper and Word

Embedding preparation

Once the exported Core ML model is imported in a client app to predict

the intent from a new utterance, it will need to be encoded using the

same one-hot embedding logic used for preparing the training dataset.