Figure 1: Overview of the RC classifier.
This library allows to quickly implement different architectures for time series data based on Reservoir Computing (RC), the family of approaches popularized in machine learning by Echo State Networks. This library is primarly design to perform classification and clustering of both univariate and multivariate time series. However, it can also be used to perform time series forecasting.
The recommended installation is with pip:
pip install reservoir-computingAlternatively, you can install the library from source:
git clone https://github.com/FilippoMB/Time-series-classification-and-clustering-with-Reservoir-Computing.git
cd Time-series-classification-and-clustering-with-Reservoir-Computing
pip install -e .The following scripts provide minimalistic examples that illustrate how to use the library for different tasks.
To run them, download the project and cd to the root folder:
git clone https://github.com/FilippoMB/Time-series-classification-and-clustering-with-Reservoir-Computing.git
cd Time-series-classification-and-clustering-with-Reservoir-ComputingClassification
python examples/classification_example.pyYou can also view the notebook or
Clustering
python examples/clustering_example.pyYou can also view the notebook or
Forecasting
python examples/forecasting_example.pyYou can also view the notebook or
In the following, we present the three main functionalities of this library.
Referring to Figure 1, the RC classifier consists of four different modules.
- The reservoir module specifies the reservoir configuration (e.g., bidirectional, leaky neurons, circle topology). Given a multivariate time series
$\mathbf{X}$ it generates a sequence of the same length of Reservoir states$\mathbf{H}$ . - The dimensionality reduction module (optionally) applies a dimensionality reduction on the sequence of the reservoir's states
$\mathbf{H}$ generating a new sequence$\mathbf{\bar H}$ . - The representation generates a vector
$\mathbf{r}_\mathbf{X}$ from the sequence of reservoir's states, which represents in vector form the original time series$\mathbf{X}$ . - The readout module is a classifier that maps the representation
$\mathbf{r}_\mathbf{X}$ into the class label$\mathbf{y}$ , associated with the time series$\mathbf{X}$ .
This library implements also the reservoir model space, a very powerful representation
The class RC_model contained in modules.py permits to specify, train and test an RC-model.
Several options are available to customize the RC model, by selecting different configurations for each module.
The training and test function requires in input training and test data, which must be provided as multidimensional numpy arrays of shape [N,T,V], with:
- N = number of samples
- T = number of time steps in each sample
- V = number of variables in each sample
Training and test labels (Ytr and Yte) must be provided in one-hot encoding format, i.e. a matrix [N,C], where C is the number of classes.
from reservoir_computing.modules import RC_model
clf = RC_model()
clf.fit(Xtr, Ytr) # Training
Yhat = clf.predict(Xte) # PredictionThe representation RC_model used for classification can be configured to directly return the time series representations, which can be used in unsupervised tasks such as clustering and dimensionality reduction.
As in the case of classification, the data must be provided as multidimensional NumPy arrays of shape [N,T,V]
from reservoir_computing.modules import RC_model
clst = RC_model(readout_type=None)
clst.fit(X)
rX = clst.input_repr # representations of the input dataThe representations rX can be used to perfrom clustering using traditional clustering algorithms for vectorial data, such as those here.
The sequences
The class RC_forecaster contained in modules.py permits to specify, train and test an RC-model for time series forecasting.
from reservoir_computing.modules import RC_forecaster
fcst = RC_forecaster()
fcst.fit(Xtr, Ytr) # Training
Yhat = fcst.predict(Xte) # PredictionsHere, Xtr, Ytr are current and future values, respectively, used for training.
The following notebooks illustrate more advanced use-cases.
- Perform dimensionality reduction, cluster analysis, and visualize the results: view or
- Probabilistic forecasting with advanced regression models as readout: view or
- Use advanced classifiers as readout: view or
There are several datasets available to perform time series classification/clustering and forecasting.
Classification and clustering
from reservoir_computing.datasets import ClfLoader
downloader = ClfLoader()
downloader.available_datasets(details=True) # Print available datasets
Xtr, Ytr, Xte, Yte = downloader.get_data('Libras') # Download dataset and return dataForecasting
Real-world time series
from reservoir_computing.datasets import PredLoader
downloader = PredLoader()
downloader.available_datasets(details=False) # Print available datasets
X = downloader.get_data('CDR') # Download dataset and return dataSynthetic time series
from reservoir_computing.datasets import SynthLoader
synth = SynthLoader()
synth.available_datasets() # Print available datasets
Xs = synth.get_data('Lorenz') # Generate synthetic time seriesPlease, consider citing the original paper if you are using this library in your reasearch
@article{bianchi2020reservoir,
title={Reservoir computing approaches for representation and classification of multivariate time series},
author={Bianchi, Filippo Maria and Scardapane, Simone and L{\o}kse, Sigurd and Jenssen, Robert},
journal={IEEE Transactions on Neural Networks and Learning Systems},
year={2020},
publisher={IEEE}
}The code is released under the MIT License. See the attached LICENSE file.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent