It is recommended to develop using a pipenv.
A development environment can be set up automatically:
$ pipenv install --dev --pre
You can work directly inside the development environment:
$ pipenv shell
Save BotTorch_Data-Training.csv
to your working directory.
To generate a model file:
$ bottorch tune --samples BotTorch_Data-Training.csv
This will generate a model.pth
file.
To make a prediction after a model has been created (model.pth
is used by default):
$ bottorch predict Icewave Chomp --samples BotTorch_Data-Training.csv
To make a ranking of all competitors, assuming competitors are saved in competitors.csv
with one competitor name per line:
$ bottorch rank --samples BotTorch_Data-Training.csv --competitors competitors.csv
Assuming the competitors file is in order of first seed to last seed, you can simulate a single elimination bracket:
$ bottorch bracket --samples BotTorch_Data-Training.csv --competitors competitors.csv
Limited "advanced" tuning parameters are available. --l1
and --l2
specify the sizes of the first two layers of the neural network. Their "ideal" size can be determined with hypertuning. Hypertuning also uses a configurable --step_size
which determines the step size used for parameter optimization, larger step sizes are faster. There is also --epochs
which can be used for number of tuning epochs run at each step. More epochs takes more time. For example:
$ bottorch hypertune --step_size 1000 --epochs 5 --samples BotTorch_Data-Training.csv
Will print "best" l1 and l2 parameters, for example Done! Best accuracy 58.7%, L1: 2032, L2: 1032
. To generate a model file (model.pth
) using the parameters:
$ bottorch tune --l1 2032 --l2 1032 --epochs 1000 --samples BotTorch_Data-Training.csv