search

JoeyAndres / rl

Reinforcement Learning library
2 stars 0 forks source link

Gradient Descent modules failing in mac. #10

Closed JoeyAndres closed 8 years ago

JoeyAndres commented 8 years ago 
-------------------------------------------------------------------------------
Scenario: Q-learning Eligibility Traces and Gradient Descent converge to a
          solution
     Given: A Mountain Car environment
      When: We do multiple episodes
      Then: At the end, we solve the Mountain Car environment in 100 iteration
-------------------------------------------------------------------------------
/Users/travis/build/JoeyAndres/rl/test/src/QLearningETGD_test.cpp:34
...............................................................................
/Users/travis/build/JoeyAndres/rl/test/src/QLearningETGD_test.cpp:78: FAILED:
  REQUIRE( iterationCount <= 100 )
with expansion:
  144 <= 100
-------------------------------------------------------------------------------
Scenario: Sarsa Eligibility Traces and Gradient Descent converge to a solution
     Given: A Mountain Car environment
      When: We do multiple episodes
      Then: At the end, we solve the Mountain Car environment in 100 iteration
-------------------------------------------------------------------------------
/Users/travis/build/JoeyAndres/rl/test/src/SarsaETGD_test.cpp:35
...............................................................................
/Users/travis/build/JoeyAndres/rl/test/src/SarsaETGD_test.cpp:62: FAILED:
  REQUIRE( iterationCount <= 100 )
with expansion:
  150 <= 100
===============================================================================
test cases: 14 | 12 passed | 2 failed
assertions: 32 | 30 passed | 2 failed
JoeyAndres commented 8 years ago

Double precision problem in osx or gcc-4.2.1