CNTCClassifier

Contextual Time-series Neural Classifier (CNTC), as described in [1].

Quickstart

from sktime.classification.deep_learning.cntc import CNTCClassifier

estimator = CNTCClassifier(n_epochs=2000, batch_size=16, filter_sizes=(16, 8), kernel_sizes=(1, 1), rnn_size=64, lstm_size=8, dense_size=64, dropout=(0.2, 0.2, 0.1, 0.1, 0.1, 0.1, 0.1), callbacks=None, verbose=False, loss='categorical_crossentropy', metrics=None, random_state=0, activation='softmax', activation_attention='sigmoid', activation_hidden='relu')

Parameters(17)

activationstring or a tf callable, default=”softmax”

Activation function used in the output layer. List of available activation functions: https://keras.io/api/layers/activations/

activation_attentionstring, default = “sigmoid”

Activation function inside the self attention module; List of available keras activation functions: https://keras.io/api/layers/activations/

activation_hiddenstring or a tf callable, default=”relu”

Activation function used in the hidden layers. List of available activation functions: https://keras.io/api/layers/activations/

n_epochsint, default = 2000

the number of epochs to train the model

batch_sizeint, default = 16

the number of samples per gradient update.

filter_sizestuple of shape (2), default = (16, 8)

filter sizes for CNNs in CCNN arm.

kernel_sizestwo-tuple, default = (1, 1)

the length of the 1D convolution window for CNNs in CCNN arm.

rnn_sizeint, default = 64

number of rnn units in the CCNN arm.

lstm_sizeint, default = 8

number of lstm units in the CLSTM arm.

dense_sizeint, default = 64

dimension of dense layer in CNTC.

dropoutfloat or tuple of floats, default = (0.2, 0.2, 0.1, 0.1, 0.1, 0.1, 0.1)

dropout rate(s), in the range [0, 1). If a single float is provided, the same dropout rate is applied to all layers. If a tuple is provided, it should have 7 values corresponding to: (conv1_dropout, rnn1_dropout, conv2_dropout, lstm_dropout,

avg_dropout, att_dropout, mlp_dropout)

where mlp_dropout is applied to both MLP layers.

random_stateint or None, default=None

Seed for random number generation.

verboseboolean, default = False

whether to output extra information

lossstring, default=”mean_squared_error”

fit parameter for the keras model

optimizerkeras.optimizer, default=keras.optimizers.Adam(),

metricslist of strings, default=[“accuracy”],

callbackslist of keras.callbacks, default = None,

Examples

>>> from sktime.classification.deep_learning.cntc import CNTCClassifier
>>> from sktime.datasets import load_unit_test
>>> X_train, y_train = load_unit_test (split = "train", return_X_y = True)
>>> X_test, y_test = load_unit_test (split = "test", return_X_y = True)
>>> cntc = CNTCClassifier ()
>>> cntc. fit (X_train, y_train) CNTCClassifier(
... )

References

1. [1 ] Network originally defined in: @article{FULLAHKAMARA202057, title = {Combining contextual neural networks for time series classification}, journal = {Neurocomputing}, volume = {384}, pages = {57-66}, year = {2020}, issn = {0925-2312}, doi = { https://doi.org/10.1016/j.neucom.2019.10.113 }, url = { https://www.sciencedirect.com/science/article/pii/S0925231219316364 }, author = {Amadu {Fullah Kamara} and Enhong Chen and Qi Liu and Zhen Pan}, keywords = {Time series classification, Contextual convolutional neural networks, Contextual long short-term memory, Attention, Multilayer perceptron}, }