gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-04-07 05:33:31 +02:00
Code Issues Projects Releases Wiki Activity

Create a dataset using an agent (#306)

Browse Source 
Generate a dataset using an agent (allowing to select between this and a random dataset)
This commit is contained in:
Gal Leibovich
2019-05-28 09:34:49 +03:00
committed by GitHub
parent 342b7184bc
commit 9e9c4fd332
26 changed files with 351 additions and 111 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
rl_coach/tests/exploration_policies/test_e_greedy.py
View File

@@ -21,14 +21,14 @@ def test_get_action():
# verify that test phase gives greedy actions (evaluation_epsilon = 0)
policy.change_phase(RunPhase.TEST)
for i in range(100):
best_action = policy.get_action(np.array([10, 20, 30]))
best_action, _ = policy.get_action(np.array([10, 20, 30]))
assert best_action == 2
# verify that train phase gives uniform actions (exploration = 1)
policy.change_phase(RunPhase.TRAIN)
counters = np.array([0, 0, 0])
for i in range(30000):
best_action = policy.get_action(np.array([10, 20, 30]))
best_action, _ = policy.get_action(np.array([10, 20, 30]))
counters[best_action] += 1
assert np.all(counters > 9500) # this is noisy so we allow 5% error