initial implementation of NeuralModel

bibmon/_neural_model.py

@@ -0,0 +1,71 @@
+from ._generic_model import GenericModel
+
+from keras.models import Sequential
+from keras.layers import LSTM, Dense, Dropout
+from keras.optimizers import Adam
+
+###############################################################################
+
+class NeuralModel(GenericModel):
+    """
+    Model that uses Keras to apply Deep Learning to find anomaly.
+
+    Parameters
+    ----------
+    outputCount: int
+        The quantity of possible states for the process (normal, anomaly, possible anomaly, etc)
+    columCount: int
+        The number of columns/features this model can handle
+    lstmshapes: list, optional
+        Whether permutation variable importance should be calculated.
+    dropout: float, optional
+        The dropout of each LSTM layer.
+        """     
+
+    ###########################################################################
+
+    def __init__ (self, columCount, outputCount,
+                  lstmshapes=[128, 64], denseshapes=[16, 32], dropout = 0.2):
+
+        self.model = Sequential()
+
+        self.model.add(LSTM(lstmshapes[0], stateful=True, return_sequences=True, batch_input_shape=(1, 1, columCount)))
+        self.model.add(Dropout(dropout))
+
+        for shape in lstmshapes[1:-1]:
+            self.model.add(LSTM(shape, stateful=True, return_sequences=True))
+            self.model.add(Dropout(dropout))
+
+        if len(lstmshapes) > 1:
+            self.model.add(LSTM(lstmshapes[-1], stateful=True, return_sequences=False))
+            self.model.add(Dropout(dropout))
+
+        for shape in denseshapes:
+            self.model.add(Dense(shape, activation='relu'))
+
+        self.model.add(Dense(outputCount, activation='softmax'))
+
+        optimizer = Adam(learning_rate=0.001)
+        self.model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])
+
+    ###########################################################################
+
+    def train_core (self):
+        self.model.fit(
+            self.X_train.values,
+            self.Y_train.values.squeeze(),
+            epochs=20,
+            batch_size=64,
+            validation_split=0.2
+        )
+
+    ###########################################################################
+
+    def map_from_X(self, X):
+        return self.model.predict(X)
+
+    ###########################################################################
+
+    def set_hyperparameters (self, params_dict):   
+        for key, value in params_dict.items():
+            setattr(self.regressor, key, value)