$ python3 main.py
# Replay memory minimum and maximum size
MIN_MEM_SIZE, MAX_MEM_SIZE = 2400, 80000
# Batch size for NN ingestion
BATCH_SIZE = 32
# Sequence length for NN ingestion
SEQUENCE_LENGTH = 8
# Number of states to ignore when computing loss
IGNORE_UP_TO = 4
# Maximum episode duration, in frames
MAX_EPISODE_LENGTH = 125 # 500 with a frame skip of 4
# Number of training steps
TRAINING_STEPS = 1000
# Number of backpropagation steps to execute after each episode
BACKPROP_STEPS = 15
# Number of training steps
QLEARNING_STEPS = 5000
# Number of steps during which epsilon should be decreased
GREEDY_STEPS = 1000
# Maximum number of cores to use
MAX_CPUS = 32
# Number of possible actions
N_ACTIONS = 3
# Learning rate for tensorflow optimizers
LEARNING_RATE = 0.001
# Use the game features in the learning phase
USE_GAME_FEATURES = False
# Use LSTM or simple DQN
USE_RECURRENCE = True
# Learn Q in the learning phase
LEARN_Q = True
DEATH_PENALTY = 25
KILL_REWARD = 100
PICKUP_REWARD = 4
try:
from local_config import * # NOQA
except ImportError:
pass
cat: local_config.py: No such file or directory
commit 6d67d0397c9a277d780003d8d64199630fa70407
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
Building main DRQN
Learn Q only
Building target DRQN
Traceback (most recent call last):
File "/Users/qiuwei/Research/SYS-CMU/deepdoom/src/agent.py", line 151, in init_phase
saver = tf.train.import_meta_graph('model.ckpt.meta')
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/training/saver.py", line 1570, in import_meta_graph
meta_graph_def = meta_graph.read_meta_graph_file(meta_graph_or_file)
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/framework/meta_graph.py", line 402, in read_meta_graph_file
raise IOError("File %s does not exist." % filename)
OSError: File model.ckpt.meta does not exist.
=== Recreate new model ! ===
---------
:: bootstrap_phase ::
mem_size,n_games
165,3
415,5
587,9
892,13
1226,17
1410,20
1740,23
1896,26
2120,29
ViZDoom ERROR
2293,32
2465,35
---------
:: learning_phase ::
qlearning_step,epsilon,reward,steps,loss_Q,loss_gf,Q0,Q1,Q2
^C^C
The chkpt file doesn't exist when I run the code
OS: MacOS