matthewmcleod / curiosity Goto Github PK

• Mountain Car experiment
• Reproduce the TMaze experiments (linear/Tabular FA)
• 1d TMaze environment

• TD(\lambda)
• ESARSA(\lambda) -> Apply new optimizer functionality
• LSTD
• Actor Critic (behavior)
• Successor Representation
• GPI (variants)

• TB - QLearners
• TB - SRLearners

Move everything global RNG, make life easy matt schleg...

Initial Auto code done by other Matt. We should think about redesign and splitting dependencies.

• Need to add savefile functionality. But will do after logger.

This is to keep track of progress for the mountain car project:

• Decide whether to redesign auto #6
• Integrate Mountain Car environment
• Port over monte carlo code. Parallelize??>>>
• Behaviour learning
• Get experiment up and running to be able to test
• Figure out what the TC should be (i.e. num tiles, tilings etc...)
• Specialize some GVFs and ensure there are no obvious errors when running (i.e. exceptions)
• Run quick test to make sure things are ok.
• Finalize sweep parameters