[ENH] Add and Validate n_layers, n_units, activation & dropout_rate kwargs to MLPNetwork

aeon/classification/deep_learning/_mlp.py

@@ -21,6 +21,17 @@ class MLPClassifier(BaseDeepClassifier):
 
     Parameters
     ----------
+    n_layers : int, optional (default=3)
+        The number of dense layers in the MLP.
+    n_units : Union[int, List[int]], optional (default=200)
+        Number of units in each dense layer.
+    activation : Union[str, List[str]], optional (default='relu')
+        Activation function(s) for each dense layer.
+    dropout_rate : Union[float, List[Union[int, float]]], optional (default=None)
+        Dropout rate(s) for each dense layer. If None, a default rate of 0.2 is used.
+        Dropout rate(s) are typically a number in the interval [0, 1].
+    dropout_last : float, default = 0.3
+        The dropout rate of the last layer.
     use_bias : bool, default = True
         Condition on whether or not to use bias values for dense layers.
     n_epochs : int, default = 2000
@@ -76,10 +87,6 @@ class MLPClassifier(BaseDeepClassifier):
         a single string metric is provided, it will be
         used as the only metric. If a list of metrics are
         provided, all will be used for evaluation.
-    activation : string or a tf callable, default="sigmoid"
-        Activation function used in the output linear layer.
-        List of available activation functions:
-        https://keras.io/api/layers/activations/
 
     Notes
     -----
@@ -104,6 +111,11 @@ class MLPClassifier(BaseDeepClassifier):
 
     def __init__(
         self,
+        n_layers=3,
+        n_units=200,
+        activation="relu",
+        dropout_rate=None,
+        dropout_last=None,
         use_bias=True,
         n_epochs=2000,
         batch_size=16,
@@ -120,16 +132,19 @@ def __init__(
         last_file_name="last_model",
         init_file_name="init_model",
         random_state=None,
-        activation="sigmoid",
         optimizer=None,
     ):
+        self.n_layers = n_layers
+        self.n_units = n_units
+        self.activation = activation
+        self.dropout_rate = dropout_rate
+        self.dropout_last = dropout_last
         self.callbacks = callbacks
         self.n_epochs = n_epochs
         self.verbose = verbose
         self.loss = loss
         self.metrics = metrics
         self.use_mini_batch_size = use_mini_batch_size
-        self.activation = activation
         self.use_bias = use_bias
         self.file_path = file_path
         self.save_best_model = save_best_model
@@ -147,7 +162,14 @@ def __init__(
             last_file_name=last_file_name,
         )
 
-        self._network = MLPNetwork(use_bias=self.use_bias)
+        self._network = MLPNetwork(
+            n_layers=self.n_layers,
+            n_units=self.n_units,
+            activation=self.activation,
+            dropout_rate=self.dropout_rate,
+            dropout_last=self.dropout_last,
+            use_bias=self.use_bias,
+        )
 
     def build_model(self, input_shape, n_classes, **kwargs):
         """Construct a compiled, un-trained, keras model that is ready for training.

aeon/networks/_mlp.py

@@ -3,6 +3,10 @@
 __maintainer__ = ["hadifawaz1999"]
 
 
+import typing
+
+import numpy as np
+
 from aeon.networks.base import BaseDeepLearningNetwork
 
 
@@ -13,6 +17,17 @@ class MLPNetwork(BaseDeepLearningNetwork):
 
     Parameters
     ----------
+    n_layers : int, optional (default=3)
+        The number of dense layers in the MLP.
+    n_units : Union[int, List[int]], optional (default=200)
+        Number of units in each dense layer.
+    activation : Union[str, List[str]], optional (default='relu')
+        Activation function(s) for each dense layer.
+    dropout_rate : Union[float, List[Union[int, float]]], optional (default=None)
+        Dropout rate(s) for each dense layer. If None, a default rate of 0.2 is used.
+        Dropout rate(s) are typically a number in the interval [0, 1].
+    dropout_last : float, default = 0.3
+        The dropout rate of the last layer.
     use_bias : bool, default = True
         Condition on whether or not to use bias values for dense layers.
 
@@ -35,12 +50,22 @@ class MLPNetwork(BaseDeepLearningNetwork):
 
     def __init__(
         self,
-        use_bias=True,
+        n_layers: int = 3,
+        n_units: typing.Union[int, list[int]] = 200,
+        activation: typing.Union[str, list[str]] = "relu",
+        dropout_rate: typing.Union[float, list[float]] = None,
+        dropout_last: float = None,
+        use_bias: bool = True,
     ):
-        self.use_bias = use_bias
-
         super().__init__()
 
+        self.n_layers = n_layers
+        self.n_units = n_units
+        self.activation = activation
+        self.dropout_rate = dropout_rate
+        self.dropout_last = dropout_last
+        self.use_bias = use_bias
+
     def build_network(self, input_shape, **kwargs):
         """Construct a network and return its input and output layers.
 
@@ -54,27 +79,80 @@ def build_network(self, input_shape, **kwargs):
         input_layer : a keras layer
         output_layer : a keras layer
         """
+        if isinstance(self.activation, str):
+            self._activation = [self.activation] * self.n_layers
+        elif isinstance(self.activation, list):
+            assert (
+                len(self.activation) == self.n_layers
+            ), "There should be an `activation` function associated with each layer."
+            assert all(
+                isinstance(a, str) for a in self.activation
+            ), "Activation must be a list of strings."
+            self._activation = self.activation
+
+        if self.dropout_rate is None:
+            self._dropout_rate = [0.1]
+            self._dropout_rate.extend([0.2] * (self.n_layers - 1))
+            assert np.all(
+                np.array(self._dropout_rate) - 1 <= 0
+            ), "Dropout rate(s) should be in the interval [0, 1]."
+        elif isinstance(self.dropout_rate, (int, float)):
+            self._dropout_rate = [float(self.dropout_rate)] * self.n_layers
+            assert np.all(
+                np.array(self._dropout_rate) - 1 <= 0
+            ), "Dropout rate(s) should be in the interval [0, 1]."
+        elif isinstance(self.dropout_rate, list):
+            assert (
+                len(self.dropout_rate) == self.n_layers
+            ), "There should be a `dropout_rate` associated with each layer."
+            assert all(
+                isinstance(d, (int, float)) for d in self.dropout_rate
+            ), "Dropout rates must be int or float."
+            assert (
+                len(self.dropout_rate) == self.n_layers
+            ), "Dropout list length must match number of layers."
+            self._dropout_rate = [float(d) for d in self.dropout_rate]
+            assert np.all(
+                np.array(self._dropout_rate) - 1 <= 0
+            ), "Dropout rate(s) should be in the interval [0, 1]."
+
+        if isinstance(self.n_units, int):
+            self._n_units = [self.n_units] * self.n_layers
+        elif isinstance(self.n_units, list):
+            assert all(
+                isinstance(u, int) for u in self.n_units
+            ), "`n_units` must be int for all layers."
+            assert (
+                len(self.n_units) == self.n_layers
+            ), "`n_units` length must match number of layers."
+            self._n_units = self.n_units
+
+        if self.dropout_last is None:
+            self._dropout_last = 0.3
+        else:
+            assert isinstance(self.dropout_last, float) or (
+                int(self.dropout_last // 1) in [0, 1]
+            ), "a float is expected in the `dropout_last` argument."
+            assert (
+                self.dropout_last - 1 <= 0
+            ), "`dropout_last` argument must be a number in the interval [0, 1]"
+            self._dropout_last = self.dropout_last
+
         from tensorflow import keras
 
-        # flattened because multivariate should be on same axis
         input_layer = keras.layers.Input(input_shape)
         input_layer_flattened = keras.layers.Flatten()(input_layer)
 
-        layer_1 = keras.layers.Dropout(0.1)(input_layer_flattened)
-        layer_1 = keras.layers.Dense(500, activation="relu", use_bias=self.use_bias)(
-            layer_1
-        )
-
-        layer_2 = keras.layers.Dropout(0.2)(layer_1)
-        layer_2 = keras.layers.Dense(500, activation="relu", use_bias=self.use_bias)(
-            layer_2
-        )
+        x = input_layer_flattened
 
-        layer_3 = keras.layers.Dropout(0.2)(layer_2)
-        layer_3 = keras.layers.Dense(500, activation="relu", use_bias=self.use_bias)(
-            layer_3
-        )
+        for idx in range(0, self.n_layers):
+            x = keras.layers.Dropout(self._dropout_rate[idx])(x)
+            x = keras.layers.Dense(
+                self._n_units[idx],
+                activation=self._activation[idx],
+                use_bias=self.use_bias,
+            )(x)
 
-        output_layer = keras.layers.Dropout(0.3)(layer_3)
+        output_layer = keras.layers.Dropout(self._dropout_last)(x)
 
         return input_layer, output_layer

aeon/regression/deep_learning/_mlp.py

@@ -21,6 +21,17 @@ class MLPRegressor(BaseDeepRegressor):
 
     Parameters
     ----------
+    n_layers : int, optional (default=3)
+        The number of dense layers in the MLP.
+    n_units : Union[int, List[int]], optional (default=200)
+        Number of units in each dense layer.
+    activation : Union[str, List[str]], optional (default='relu')
+        Activation function(s) for each dense layer.
+    dropout_rate : Union[float, List[Union[int, float]]], optional (default=None)
+        Dropout rate(s) for each dense layer. If None, a default rate of 0.2 is used.
+        Dropout rate(s) are typically a number in the interval [0, 1].
+    dropout_last : float, default = 0.3
+        The dropout rate of the last layer.
     use_bias : bool, default = True
         Condition on whether or not to use bias values for dense layers.
     n_epochs : int, default = 2000
@@ -72,10 +83,6 @@ class MLPRegressor(BaseDeepRegressor):
         by `np.random`.
         Seeded random number generation can only be guaranteed on CPU processing,
         GPU processing will be non-deterministic.
-    activation : string or a tf callable, default="relu"
-        Activation function used in the output linear layer.
-        List of available activation functions:
-        https://keras.io/api/layers/activations/
     output_activation : str = "linear"
         Activation for the last layer in a Regressor.
     optimizer : keras.optimizer, default = tf.keras.optimizers.Adam()
@@ -100,6 +107,11 @@ class MLPRegressor(BaseDeepRegressor):
 
     def __init__(
         self,
+        n_layers=3,
+        n_units=200,
+        activation="relu",
+        dropout_rate=None,
+        dropout_last=None,
         use_bias=True,
         n_epochs=2000,
         batch_size=16,
@@ -115,16 +127,19 @@ def __init__(
         last_file_name="last_model",
         init_file_name="init_model",
         random_state=None,
-        activation="relu",
         output_activation="linear",
         optimizer=None,
     ):
+        self.n_layers = n_layers
+        self.n_units = n_units
+        self.activation = activation
+        self.dropout_rate = dropout_rate
+        self.dropout_last = dropout_last
         self.callbacks = callbacks
         self.n_epochs = n_epochs
         self.verbose = verbose
         self.loss = loss
         self.metrics = metrics
-        self.activation = activation
         self.use_bias = use_bias
         self.file_path = file_path
         self.save_best_model = save_best_model
@@ -143,7 +158,14 @@ def __init__(
             last_file_name=last_file_name,
         )
 
-        self._network = MLPNetwork(use_bias=self.use_bias)
+        self._network = MLPNetwork(
+            n_layers=self.n_layers,
+            n_units=self.n_units,
+            activation=self.activation,
+            dropout_rate=self.dropout_rate,
+            dropout_last=self.dropout_last,
+            use_bias=self.use_bias,
+        )
 
     def build_model(self, input_shape, **kwargs):
         """Construct a compiled, un-trained, keras model that is ready for training.