def test_keras_evaluate(self):
keras_model = _create_test_cnn_model()
sample_data = [
collections.OrderedDict(x=np.ones([1, 28, 28, 1],
dtype=np.float32),
y=np.ones([1], dtype=np.int32))
]
metrics = [tf.keras.metrics.SparseCategoricalAccuracy()]
metrics = dp_fedavg.keras_evaluate(keras_model, sample_data, metrics)
accuracy = metrics[0].result()
self.assertIsInstance(accuracy, tf.Tensor)
self.assertBetween(accuracy, 0.0, 1.0)
def test_tff_learning_evaluate(self):
it_process = dp_fedavg.build_federated_averaging_process(
_tff_learning_model_fn)
server_state = it_process.initialize()
sample_data = [
collections.OrderedDict(x=np.ones([1, 28, 28, 1],
dtype=np.float32),
y=np.ones([1], dtype=np.int32))
]
keras_model = _create_test_cnn_model()
server_state.model.assign_weights_to(keras_model)
sample_data = [
collections.OrderedDict(x=np.ones([1, 28, 28, 1],
dtype=np.float32),
y=np.ones([1], dtype=np.int32))
]
metrics = [tf.keras.metrics.SparseCategoricalAccuracy()]
metrics = dp_fedavg.keras_evaluate(keras_model, sample_data, metrics)
accuracy = metrics[0].result()
self.assertIsInstance(accuracy, tf.Tensor)
self.assertBetween(accuracy, 0.0, 1.0)
def evaluate_fn(model_weights, dataset):
model.from_weights(model_weights)
metrics = dp_fedavg.keras_evaluate(model.keras_model, dataset,
keras_metics)
return collections.OrderedDict(
(metric.name, metric.result().numpy()) for metric in metrics)
