To run the code, please install tqdm, pytorch, scipy, and matplotlib libraries.
To run the code, please execute main.py from the directory where it is located
with at least one of the following flags:
--hyper, --long, --curriculum, --transfer, --figures.
Each flag will trigger a portion of the experiments as described in the doc,
which can be accessed by running main.py with the -h flag.
--figures flag triggers reproduction of the figures.

The results of each experiment are stored in results/{EXPERIMENT} folder, where
{EXPERIMENT} corresponds to one of the flags above. Note that the code will
always attempt loading and reporting the results stored in the results directory.
To force-reproduce all experiments from scratch, remove the results directory.
Running all experiments will take several days.

Recreating curriculum/meta_20_2.txt, which is the file characterising each
randomly generated curricula, and especially results/hyper directory, which stores the
results of the hyperparameter search, can make some of the figures irreproducable
as they depend on hyperparameters and curricula specifications. Removal of anything else
should not affect the ability of the code to reproduce the figures (but should
and will change their content due to randomness in NN weight initialization) so long as
the results are recreated after removal by running main.py with the appropriate flag.