Coverage for src/sensai/tensorflow/tf_mlp.py: 37%

1from tensorflow import keras 

2 

3from .tf_base import KerasVectorRegressionModel 

4from .. import normalisation 

5 

6 

7class KerasMultiLayerPerceptronVectorRegressionModel(KerasVectorRegressionModel): 

8 def __init__(self, hidden_dims=(5,5), hidden_activation="sigmoid", output_activation="sigmoid", loss="mse", 

9 metrics=("mse",), optimiser="adam", normalisation_mode=normalisation.NormalisationMode.MAX_BY_COLUMN, **kwargs): 

10 super().__init__(normalisation_mode, loss, metrics, optimiser, **kwargs) 

11 self.hiddenDims = hidden_dims 

12 self.hiddenActivation = hidden_activation 

13 self.outputActivation = output_activation 

14 

15 def __str__(self): 

16 params = dict(hiddenDims=self.hiddenDims, hiddenActivation=self.hiddenActivation, 

17 outputActivation=self.outputActivation) 

18 return f"{self.__class__.__name__}{params}={super().__str__()}" 

19 

20 def _create_model(self, input_dim, output_dim): 

21 model_inputs = keras.Input(shape=(input_dim,), name='input') 

22 x = model_inputs 

23 for i, hiddenDim in enumerate(self.hiddenDims): 

24 x = keras.layers.Dense(hiddenDim, activation=self.hiddenActivation, name='dense_%d' % i)(x) 

25 model_outputs = keras.layers.Dense(output_dim, activation=self.outputActivation, name='predictions')(x) 

26 return keras.Model(inputs=model_inputs, outputs=model_outputs) 

27 

28