Here we maintain examples of Python scripts which show how to use hdlib for building vector-symbolic architectures.
The chopin2.py script is a reimplementation of chopin2 which makes use of hdlib for building a supervised classification model following the hyperdimensional computing paradigm.
chopin2 has been originally proposed for dealing with massive amount of samples and features and has been applied on DNA-Methylation data of cancer.
Cumbo F, Cappelli E, Weitschek E. A brain-inspired hyperdimensional computing approach for classifying massive dna methylation data of cancer. Algorithms. 2020 Sep 17;13(9):233. https://doi.org/10.3390/a13090233
As the original software, the chopin2.py script only accepts numerical datasets in input. The first row must contain the header with the name of the features, while the first column contains the name of the samples and the last column the class labels.
Note: Given an input matrix M, values in position M(ij) must be numbers, except for the first row, first column, and last column.
Here is a list of available options:
| Option | Default | Mandatory | Description |
|---|---|---|---|
--input |
⚑ | Path to the input matrix | |
--fieldsep |
\t |
Input field separator | |
--dimensionality |
10000 |
Vectors dimensionality | |
--levels |
⚑ | Number of level vectors | |
--kfolds |
0 |
Number of folds for cross-validating the classification model | |
--test-percentage |
0 |
Percentage of data points for defining the test set | |
--feature-selection |
Run the feature selection (backward or forward variable elimination) and report a ranking of features based on their importance |
||
--retrain |
0 |
Number of retraining iterations | |
--threshold |
0.6 |
Threshold on the accuracy score. Used in conjunction with --feature-selection |
|
--uncertainty |
5.0 |
Uncertainty percentage. Used in conjunction with --feature-selection |
|
--nproc |
1 |
Make it parallel when possible | |
--version |
Print the tool version and exit |
Note: The chopin2_iris.sh automatically downloads the iris.csv dataset from the scikit-learn repository on GitHub and run chopin2.py by using 10 level vectors and selecting features through the backward variable elimination technique in 5-folds cross-validation. Also note that this must be intended just as an example about how to run chopin2.py and results strongly depend on the input dataset.