SSL

What is what

sslearn contains models and everything connected with machine learning.
experimentarium contains functionality to launch some testing.
config.py parameters of the models that are to be tested.

Benchmarks

Adding new ones

One needs to properly provide datareader to experimentarium/data_react/rfuncs.py , update experimentarium/data_react/dataset2dir.json and experimentarium/data_react/dir2url.json .

Testing

Examine experementarium/run.py that supports command line arguments. Than run, supposing your current directory is SSL , command

python3 experementarium/run.py --arg1 value1 --arg2 value2 --arg3 value3

By default it takes data from ../SSL/data if there is no such folder it tries to download data from the internet.

There is also a possibility to avoid typing arguments to variety command line arguments, just create file, for instance, ../SSL/run.py with the following content (of course, run_params may vary).

import sys

repo_path = "SSL"
sys.path.append(repo_path)

from experimentarium.utils import ShellTestRunner  # noqa

if __name__ == "__main__":
    run_root = "SSL/experimentarium"
    run_params = {
        "model": ["sla"],
        "benchmarks": [
            "pendigits_4_9",
        ],
        "verbose": "False",
        "lsizes": [0.005],
        "n-states": 1,
        "debug": "True",
        "log": "False",
        "progress-bar": "True",
        "merge-results": "True",
    }
    ShellTestRunner().run(run_root, **run_params)

The only thing now is to change run_params to whatever you need — they need to be the same as in SSL/experimentarium/run.py — and execute:

python3 ../SSL/run.py

To see running parameters and their defaults, execute:

python3 SSL/experimentarium/run.py --help

NOTE: When running clean experiments use default values for --lsizes abd --n-states .

SSL-GRENOBLE/SSL

SSL

What is what

Benchmarks

Adding new ones

Testing