QlearningExample.torch

Torch plays catch! Based on Eder Santanas' implementation in keras. Highly recommend reading his informative and easy to follow blog post here.

Agent has to catch the fruit before it falls to the ground. Agent wins if he succeeds to catch the fruit, loses if he fails.

Dependencies

To install torch7 follow the guide here.

Other dependencies can be installed via luarocks:

luarocks install optim
luarocks install image

To train a model, run the Train.lua script. You can configure parameters such as below:

th Train.lua -epoch 1000 #Configures the number of epochs. More parameters are available, check scrip.t

To visualise the agent playing the game after training a model, use the TorchPlaysCatch.lua script using qlua rather than th as below:

qlua TorchPlaysCatch.lua

Much like the train script, there are configurable options. Check the script for more details!