def train(self, dataPointsTrain, dataPointsValidation, dataAugmentation):
self.logManager.newLogSession("Creating Histogram plot: Training data")
self.plotManager.createCategoricalHistogram(dataPointsTrain, "train")
self.logManager.newLogSession(
"Creating Histogram plot: Validation data")
self.plotManager.createCategoricalHistogram(dataPointsValidation,
"validation")
self.logManager.newLogSession("Training Model")
optimizer = Adagrad()
self.optimizerType = "AdaGrad"
if not self.logManager is None:
self.logManager.write("Training Strategy: " +
str(optimizer.get_config()))
self.logManager.write("--- Training Optimizer: " +
str(self.optimizerType))
self.logManager.write("--- Training Strategy: " +
str(optimizer.get_config()))
self.logManager.write("--- Training Batchsize: " +
str(self.batchSize))
self.logManager.write("--- Training Number of Epochs: " +
str(self.numberOfEpochs))
self.model.compile(loss="categorical_crossentropy",
optimizer=optimizer,
metrics=[
'accuracy', 'categorical_accuracy',
metrics.fbeta_score, metrics.recall,
metrics.precision
])
filepath = self.experimentManager.modelDirectory + "/weights.best.hdf5"
checkpoint = ModelCheckpoint(filepath,
monitor='val_categorical_accuracy',
verbose=1,
save_best_only=True,
mode='max')
early_stopping = EarlyStopping(monitor='val_loss',
mode="min",
patience=40)
reduce_lr = ReduceLROnPlateau(factor=0.5,
monitor='val_loss',
min_lr=1e-5,
patience=2)
callbacks_list = [checkpoint, early_stopping, reduce_lr]
history_callback = self.model.fit(
dataPointsTrain.dataX,
dataPointsTrain.dataY,
batch_size=self.batchSize,
epochs=self.numberOfEpochs,
validation_data=(dataPointsValidation.dataX,
dataPointsValidation.dataY),
shuffle=True,
callbacks=callbacks_list)
if not self.logManager is None:
self.logManager.write(str(history_callback.history))
self.plotManager.createTrainingPlot(history_callback)
self.logManager.endLogSession()
def train(self, dataPointsTrain, dataPointsValidation, dataAugmentation):
self.logManager.newLogSession("Creating Histogram plot: Training data")
self.plotManager.createCategoricalHistogram(dataPointsTrain, "train")
self.logManager.newLogSession(
"Creating Histogram plot: Validation data")
self.plotManager.createCategoricalHistogram(dataPointsValidation,
"validation")
self.logManager.newLogSession("Training Model")
optimizer = Adagrad(lr=0.01, epsilon=1e-08, decay=0.0001)
self.optimizerType = "AdaGrad"
if not self.logManager is None:
self.logManager.write("Training Strategy: " +
str(optimizer.get_config()))
self.logManager.write("--- Training Optimizer: " +
str(self.optimizerType))
self.logManager.write("--- Training Strategy: " +
str(optimizer.get_config()))
self.logManager.write("--- Training Batchsize: " +
str(self.batchSize))
self.logManager.write("--- Training Number of Epochs: " +
str(self.numberOfEpochs))
from keras.losses import logcosh
self.model.compile(loss="mean_absolute_error",
optimizer=optimizer,
metrics=['mse', metrics.ccc])
filepath = self.experimentManager.modelDirectory + "/weights.best.hdf5"
checkPoint = ModelCheckpoint(
filepath,
monitor='val_arousal_output_mean_squared_error',
verbose=1,
save_best_only=True,
save_weights_only=False,
mode='auto',
period=1)
reduce_lr = ReduceLROnPlateau(
monitor='val_valence_output_mean_squared_error',
factor=0.2,
patience=5,
min_lr=0.0001,
verbose=1)
callbacks_list = [reduce_lr]
history_callback = self.model.fit(
[dataPointsTrain.dataXVideo, dataPointsTrain.dataXAudio],
[dataPointsTrain.dataY[:, 0], dataPointsTrain.dataY[:, 1]],
batch_size=self.batchSize,
epochs=self.numberOfEpochs,
validation_data=([
dataPointsValidation.dataXVideo,
dataPointsValidation.dataXAudio
], [
dataPointsValidation.dataY[:, 0], dataPointsValidation.dataY[:,
1]
]),
shuffle=True,
callbacks=callbacks_list)
if not self.logManager is None:
self.logManager.write(str(history_callback.history))
self.plotManager.createTrainingPlot(history_callback)
self.logManager.endLogSession()
