Commit 2e56ec61 authored by hanzopgp's avatar hanzopgp
Browse files

model not converging

parent 5e906ac0
......@@ -21,6 +21,9 @@ if __name__ == '__main__':
X = X.astype("float32")
y = {"value_head": y_values.astype("float32"), "value_opp_head": y_opp_values.astype("float32"), "policy_head": y_distrib.reshape(y_distrib.shape[0], -1).astype("float32")}
print(len(y["policy_head"][y["policy_head"]!=0]))
print(len(y["policy_head"][y["policy_head"]==0]))
champion_path = MODEL_PATH+GAME_NAME+"_"+"champion"+".h5"
outsider_path = MODEL_PATH+GAME_NAME+"_"+"outsider"+".h5"
......
......@@ -27,7 +27,7 @@ NUM_DOJO = 2
MAX_ITERATION_AGENTS_DOJO = 30
THINKING_TIME_AGENTS_DOJO = -1
NUM_EPISODE = 2
NUM_EPISODE = 10
MAX_ITERATION_AGENT1 = 30
MAX_ITERATION_AGENT2 = 30
THINKING_TIME_AGENT1 = -1
......@@ -60,7 +60,7 @@ MAX_SAMPLE = 5000 # can decide the max size of the dataset
######### NN parameters #########
TRAIN_SAMPLE_SIZE = 4096
TRAIN_SAMPLE_SIZE = 16000
RANDOM_SEED = 42
BATCH_SIZE = 512
VERBOSE = 1
......
......@@ -7,7 +7,7 @@ N_ROW = 8
N_COL = 8
N_LEVELS = 24
N_ADDITIONAL_FEATURES = 1 # currently only the color of the current player
N_TIME_STEP = 3 # number of past state we keep for our representation
N_TIME_STEP = 1 # number of past state we keep for our representation
N_REPRESENTATION_STACK = N_ADDITIONAL_FEATURES + (N_TIME_STEP * 2) * N_LEVELS
# Action representation
N_DISTANCE = 8 # the king can move on the whole diagonal
......
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment