min2net.model.MIN2Net_without_decoder

min2net.model.MIN2Net_without_decoder

MIN2Net_without_decoder class

Configures the model for training. Based on tf.keras.Model.

min2net.model.MIN2Net_without_decoder()

Arguments:

Arguments	Description	Default
input_shape	`tuple` of integers. (1, #time_point, #channel)	(1,400,20)
num_class	`int` number of class.	2
loss	`list` of `str` (name of objective function), objective function or tf.keras.losses.Loss instance.	[`triplet_loss(margin=1.0)`, `'sparse_categorical_crossentropy'`]
loss_weights	`list` of `float` [β1,β2] to weight the loss contributions of different model outputs.	[1., 1.]
latent_dim	`int` or `None`. If `None`, latent_dim is equal to number of channels `C` or 64 for 2- or 3-class classification, respectively.	`None`
epochs	`int` number of epochs to train the model.	200
batch_size	`int` or `None`. Number of samples per gradient update.	100
optimizer	`str` (name of optimizer) or `optimizer instance`. See tf.keras.optimizers.	Adam(beta_1=0.9, beta_2=0.999, epsilon=1e-08)
lr	`float` the start learning rate	1e-2
min_lr	`float` lower bound on the learning rate. See tf.keras.callbacks.ReduceLROnPlateau.	1e-3
factor	`float` factor by which the learning rate will be reduced. See tf.keras.callbacks.ReduceLROnPlateau.	0.25
patience	`int` number of epochs with no improvement after which learning rate will be reduced. See tf.keras.callbacks.ReduceLROnPlateau.	20
es_patience	`int` number of epochs with no improvement after which training will be stopped. See tf.keras.callbacks.EarlyStopping.	50
verbose	`0` or `1`. Verbosity mode. 0 = silent, 1 = progress bar.	1
log_path	`str` path to save model	‘logs’
model_name	`str` prefix to save model	‘MIN2Net_without_decoder’
**kwargs	Keyword argument to pass into the function for replacing the variables of the model such as `data_format`, `shuffle`, `metrics`, `monitor`, `mode`, `save_best_only`, `save_weight_only`, `seed`, and `class_balancing`	-

Build method

Build the model that group layers into an object with training and inference features.

MIN2Net_without_decoder.build()

Returns: Model (tf.keras.Model): Model object

Fit method

Fit the model according to the given training and validation data. This method was implemented based on tf.keras.Model.fit(). The model weights and logs will save at 'log_path'.

MIN2Net_without_decoder.fit(X_train, 
                            y_train, 
                            X_val, 
                            y_val)

Arguments:

Arguments	Description
X_train	`ndarray` Training EEG signals. shape (#trial, #depth, #time_point, #channel)
y_train	`ndarray` Label of training set. shape (#trial)
X_val	`ndarray` Validation EEG signals. shape (#trial, #depth, #time_point, #channel)
y_val	`ndarray` Label of validation set. shape (#trial)

Predict method

Generates output predictions & the loss value & metrics values for the model in test mode for the input samples. This method was implemented based on tf.keras.Model.predict() and tf.keras.Model.evaluate().

MIN2Net_without_decoder.predict(X_test, y_test)

Arguments	Description
X_test	`ndarray` Testing EEG signals. shape (#trial, #depth, #time_point, #channel)
y_test	`ndarray` Label of test set. shape (#trial)

Returns:

Y: dictionary of {y_true, y_pred, y_pred_decoder}
evaluation: dictionary of {loss, decoder_loss, classifier_loss, accuracy, f1_score}

Example

from min2net.model import MIN2Net_without_decoder
import numpy as np

model = MIN2Net_without_decoder(input_shape=(1, 400, 20), num_class=2, monitor='val_loss', shuffle=True)
model.fit(X_train, y_train, X_val, y_val)

Y, evaluation = model.predict(X_test, y_test)