Train a Keras model for Iris data classification and save result as JSON. Args: epochs: Number of epochs to traing the Keras model for. artifacts_dir: Directory to save the model artifacts (model topology JSON, weights and weight manifest) in. sequential: Whether to use a Keras
(epochs,
artifacts_dir,
sequential=False)
| 29 | |
| 30 | |
| 31 | def train(epochs, |
| 32 | artifacts_dir, |
| 33 | sequential=False): |
| 34 | """Train a Keras model for Iris data classification and save result as JSON. |
| 35 | |
| 36 | Args: |
| 37 | epochs: Number of epochs to traing the Keras model for. |
| 38 | artifacts_dir: Directory to save the model artifacts (model topology JSON, |
| 39 | weights and weight manifest) in. |
| 40 | sequential: Whether to use a Keras Sequential model, instead of the default |
| 41 | functional model. |
| 42 | |
| 43 | Returns: |
| 44 | Final classification accuracy on the training set. |
| 45 | """ |
| 46 | data_x, data_y = iris_data.load() |
| 47 | |
| 48 | if sequential: |
| 49 | model = keras.models.Sequential() |
| 50 | model.add(keras.layers.Dense( |
| 51 | 10, input_shape=[data_x.shape[1]], use_bias=True, activation='sigmoid', |
| 52 | name='Dense1')) |
| 53 | model.add(keras.layers.Dense( |
| 54 | 3, use_bias=True, activation='softmax', name='Dense2')) |
| 55 | else: |
| 56 | iris_x = keras.layers.Input((4,)) |
| 57 | dense1 = keras.layers.Dense( |
| 58 | 10, use_bias=True, name='Dense1', activation='sigmoid')(iris_x) |
| 59 | dense2 = keras.layers.Dense( |
| 60 | 3, use_bias=True, name='Dense2', activation='softmax')(dense1) |
| 61 | # pylint:disable=redefined-variable-type |
| 62 | model = keras.models.Model(inputs=[iris_x], outputs=[dense2]) |
| 63 | # pylint:enable=redefined-variable-type |
| 64 | model.compile(loss='categorical_crossentropy', optimizer='adam') |
| 65 | |
| 66 | model.fit(data_x, data_y, batch_size=8, epochs=epochs) |
| 67 | |
| 68 | # Run prediction on the training set. |
| 69 | pred_ys = np.argmax(model.predict(data_x), axis=1) |
| 70 | true_ys = np.argmax(data_y, axis=1) |
| 71 | final_train_accuracy = np.mean((pred_ys == true_ys).astype(np.float32)) |
| 72 | print('Accuracy on the training set: %g' % final_train_accuracy) |
| 73 | |
| 74 | tfjs.converters.save_keras_model(model, artifacts_dir) |
| 75 | |
| 76 | return final_train_accuracy |
| 77 | |
| 78 | |
| 79 | def main(): |