Just pip install -r requirements.txt. Only tested on python version 3.7.
Experiments are defined by json specification files in the specs directory. These control what hyperparameter grid to scan over in the data collection phase. Please don't alter them as they are currently set up to generate the same data as in the report.
The runner file contains the code for running the experiments. Simply type:
python runner.py and the experiment pipeline for both datasets will be run. This may take a couple hours, however; some models take quite a while to train, and each gets several runs through K-Fold Cross-Validation.
Results of the experiment will be stored in two new directories: intention and pulsar, representing the two datasets.
The analyzer file contains the code for generating plots and outputting the results. Simply run python analyzer.py after running the runner.py file, and all the data collected during the data collection phase will be used to generate the relevant graphs. These graphs will be output to the intention and pulsar directories. The analyzer will also output the final model comparison table to stdout, which represent tables 11 and 12 in the report.
To run the whole project end-to-end, just run the command:
python run.py. This will run the runner and analyzer in sequence for both datasets. Results can be examined in the created folders intention and pulsar.