Trance-0/CSE5100H3
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

part2 complete

Browse Source
This commit is contained in:
Zheyuan Wu
2025-11-02 12:55:49 -06:00
parent ac986ec69a
commit 248051db0d
36 changed files with 680 additions and 953 deletions
Download Patch File Download Diff File Expand all files Collapse all files

109
plot_result.py
View File

@@ -19,7 +19,8 @@ def tabulate_events(dpath):
tags = event_iterator.Tags()['scalars']
for it in summary_iterators:
assert it.Tags()['scalars'] == tags
print(f'comparing {it.Tags()["scalars"]} and {tags}')
# assert it.Tags()['scalars'] == tags
out = defaultdict(list)
steps = []
@@ -45,20 +46,100 @@ def tabulate_events(dpath):
if __name__ == '__main__':
# In the first graph, compare the learning curves (average return vs. number of environment steps) for the experiments running with batch size of 1000.
root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p1311')
df_tables = tabulate_events(root_dir)
for experiment_dir, df in df_tables:
# debug
print(f"{experiment_dir}:\n {df}")
plt.plot(df['Train_EnvstepsSoFar'], df['Train_AverageReturn'], label=experiment_dir)
plt.legend()
plt.xlabel('Number of Environment Steps')
plt.ylabel('Average Return')
plt.title('Learning Curves for Batch Size of 1000')
plt.show()
# root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p1311')
# df_tables = tabulate_events(root_dir)
# for experiment_dir, df in df_tables:
# # debug
# print(f"{experiment_dir}:\n {df}")
# plt.plot(df['Train_EnvstepsSoFar'], df['Train_AverageReturn'], label=experiment_dir)
# plt.legend()
# plt.xlabel('Number of Environment Steps')
# plt.ylabel('Average Return')
# plt.title('Learning Curves for Batch Size of 1000')
# plt.show()
# – In the second graph, compare the learning curves for the experiments running with
# batch size of 4000. (The large batch experiments.)
root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p1312')
# root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p1312')
# df_tables = tabulate_events(root_dir)
# for experiment_dir, df in df_tables:
# # debug
# print(f"{experiment_dir}:\n {df}")
# plt.plot(df['Train_EnvstepsSoFar'], df['Train_AverageReturn'], label=experiment_dir)
# plt.legend()
# plt.xlabel('Number of Environment Steps')
# plt.ylabel('Average Return')
# plt.title('Learning Curves for Batch Size of 4000')
# plt.show()
# Plot a learning curve for the baseline loss.
# root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p231')
# df_tables = tabulate_events(root_dir)
# for experiment_dir, df in df_tables:
# # debug
# print(f"{experiment_dir}:\n {df}")
# plt.plot(df['Train_EnvstepsSoFar'], df['Baseline_Loss'], label=experiment_dir)
# plt.legend()
# plt.xlabel('Number of Environment Steps')
# plt.ylabel('Baseline Loss')
# plt.title('Learning Curve for Baseline Loss')
# plt.show()
# Plot a learning curve for the evaluation return. You should expect to converge to the maximum reward of 500.
# root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p232')
# df_tables = tabulate_events(root_dir)
# for experiment_dir, df in df_tables:
# # debug
# print(f"{experiment_dir}:\n {df}")
# plt.plot(df['Train_EnvstepsSoFar'], df['Train_AverageReturn'], label=experiment_dir)
# plt.legend()
# plt.xlabel('Number of Environment Steps')
# plt.ylabel('Average Return')
# plt.title('Learning Curve for Average Return')
# plt.show()
# Run another experiment with a decreased number of baseline gradient steps (-bgs in command line) and/or baseline learning rate (-blr in command line). How does this affect
# (a) the baseline learning curve and
# root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p233')
# df_tables = tabulate_events(root_dir)
# for experiment_dir, df in df_tables:
# # debug
# print(f"{experiment_dir}:\n {df}")
# plt.plot(df['Train_EnvstepsSoFar'], df['Baseline_Loss'], label=experiment_dir)
# plt.legend()
# plt.xlabel('Number of Environment Steps')
# plt.ylabel('Baseline Loss')
# plt.title('Learning Curve for Baseline Loss for Batch Size of 5000 with Decreased Baseline Gradient Steps and/or Baseline Learning Rate')
# plt.show()
# (b) the performance of the policy?
# root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p233')
# df_tables = tabulate_events(root_dir)
# for experiment_dir, df in df_tables:
# # debug
# print(f"{experiment_dir}:\n {df}")
# plt.plot(df['Train_EnvstepsSoFar'], df['Train_AverageReturn'], label=experiment_dir)
# plt.legend()
# plt.xlabel('Number of Environment Steps')
# plt.ylabel('Average Return')
# plt.title('Learning Curve for Average Return for Batch Size of 5000 with Decreased Baseline Gradient Steps and/or Baseline Learning Rate')
# plt.show()
# How does the command line argument -na influence the performance? Why is that the case?
# root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p234')
# df_tables = tabulate_events(root_dir)
# for experiment_dir, df in df_tables:
# # debug
# print(f"{experiment_dir}:\n {df}")
# plt.plot(df['Train_EnvstepsSoFar'], df['Train_AverageReturn'], label=experiment_dir)
# plt.legend()
# plt.xlabel('Number of Environment Steps')
# plt.ylabel('Average Return')
# plt.title('Learning Curve for Average Return for Batch Size of 5000 with Command Line Argument -na')
# plt.show()
# HalfCheetah
root_dir = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'hw3', 'data','p24')
df_tables = tabulate_events(root_dir)
for experiment_dir, df in df_tables:
# debug
@@ -67,5 +148,5 @@ if __name__ == '__main__':
plt.legend()
plt.xlabel('Number of Environment Steps')
plt.ylabel('Average Return')
plt.title('Learning Curves for Batch Size of 4000')
plt.title('Learning Curve for Average Return for HalfCheetah')
plt.show()