Time series classifier

Time series classification is performed using convolutional neural networks to classify images generated from time series.

How to install dependencies?

Assuming that you have python3 and pip3 installed, third-party packages can be installed with:

pip3 install -r requirements.txt --upgrade --user

How to run on arbitrary data set?

The program is prepared to easily use data sets in a *.arff form that data sets at timeseriesclassification.com have.

1. Having two *.arff files named YourDataSet_TRAIN.arff and YourDataSet_TEST.arff, place them in datasets/YourDataSet/ directory.
2. Run python convert.py YourDataSet to convert *.arff files into the internal format used in our classifier.
3. Run python generate_model.py YourDataSet to generate images, train the model on the training data set, and save the model fo file.
4. Run python classify.py YourDataSet to classify images based on test data set.

Alternatively, you can run the program on a data set from timeseriesclassification.com collection with this command:

python run.py YourDataSet

Change YourDataSet to a name of some data set from the website. This script downloads the data set, decompress it, and perform all the steps above. Please note that this script might only run on Linux systems.

How to run all data sets from timeseriesclassification.com?

You can run python run_everything.py to test the classifier on every data set from the websites. The data sets are downloaded, all steps from above are performed for each data set, and accuracy results are saved to file. Please note that this script might only run on Linux systems.