Actually Added timesteps_total to in run_experiments.py; some testing code

Author: RaghuSpaceRajan

mdp_playground/scripts/run_experiments.py

@@ -2,6 +2,9 @@
 
 Takes a configuration file, experiment name and config number to run as
 optional arguments.
+
+e.g.: python mdp_playground/scripts/run_experiments.py -a 0 -n 0 -c \
+    default_config.py -e default_config
 """
 
 from __future__ import absolute_import
@@ -129,7 +132,7 @@ def main(args):
         "training.",
     )
     parser.add_argument(
-        "-t",
+        "-d",
         "--framework-dir",
         dest="framework_dir",
         action="store",
@@ -139,6 +142,15 @@ def main(args):
         "framework (e.g. Ray Rllib, Stable Baselines 3). This "
         "name will be passed to the framework.",
     )
+    parser.add_argument(
+        "-t",
+        "--timesteps-total",
+        dest="timesteps_total",
+        action="store",
+        default=None,
+        type=int,
+        help="Total number of env steps to run expt for."
+    )
     # parser.add_argument('-t', '--tune-hps', dest='tune_hps', action='store',
     #                     default=False, type=bool,
     #                     help='Used for tuning the hyperparameters that can be '
@@ -227,7 +239,9 @@ def main(args):
         )
         pp.pprint(tune_config)
 
-        if "timesteps_total" in dir(config):
+        if args.timesteps_total is not None:
+            timesteps_total = args.timesteps_total
+        elif "timesteps_total" in dir(config):
             timesteps_total = config.timesteps_total
         else:
             timesteps_total = tune_config["timesteps_total"]

tests/test_analysis_code.py

@@ -1,9 +1,8 @@
-import sys
+import sys, os
 from datetime import datetime
 import logging
 import copy
 import numpy as np
-from mdp_playground.envs.rl_toy_env import RLToyEnv
 import unittest
 import pytest
 
@@ -16,6 +15,24 @@
 
 
 class TestAnalysisCode(unittest.TestCase):
+
+    # ###TODO Enable once scipy can be upraded
+    # def test_generate_plots(self):
+
+    #     exit_code = os.system(
+    #         sys.executable
+    #         + " misc/generate_mdpp_plots.py -f misc/test_expt_list.txt -n 1"
+    #     )
+    #     assert exit_code == 0
+
+    #     from glob import glob
+    #     plot_list = glob("*.pdf")
+    #     plot_list_exp = ['rainbow_seq_del_train_final_reward_delay_episode_reward_mean_1d.pdf', 'rainbow_seq_del_train_final_reward_sequence_length_episode_reward_mean_1d.pdf', 'dqn_seq_del_train_learning_curves_episode_reward_mean.pdf', 'rainbow_seq_del_train_final_reward_mean_heat_map_episode_reward_mean.pdf', 'dqn_seq_del_train_final_reward_mean_heat_map_episode_reward_mean.pdf', 'dqn_seq_del_train_final_reward_sequence_length_episode_reward_mean_1d.pdf', 'rainbow_seq_del_train_final_reward_std_heat_map_episode_reward_mean.pdf', 'dqn_seq_del_train_final_reward_delay_episode_reward_mean_1d.pdf', 'dqn_seq_del_train_final_reward_std_heat_map_episode_reward_mean.pdf']
+
+    #     import collections
+    #     assert collections.Counter(plot_list) == collections.Counter(plot_list_exp), "Unexpected PDF file found when generating plots. Found:" + str(plot_list)
+
+
     @pytest.mark.skip(
         reason="CAVE dependencies throw ImportError: cannot import name 'StatusType'"
     )
@@ -63,3 +80,6 @@ def test_mdpp_to_cave(self):
             for i in range(2):
                 l = fh.readline()
             assert l.strip() == results_json_line_2
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_run_experiments.py

@@ -3,6 +3,8 @@
 import sys
 import os
 
+import numpy as np
+
 import logging
 
 from datetime import datetime
@@ -30,7 +32,7 @@ def test_dqn_test_expt(self):
 
         exit_code = os.system(
             sys.executable
-            + " mdp_playground/scripts/run_experiments.py -a 0 -n 0 -c experiments/dqn_test_expt.py -e dqn_test_expt"
+            + " run_experiments.py -n 0 -c experiments/dqn_test_expt.py -e dqn_test_expt"
         )
         assert exit_code == 0
 
@@ -43,7 +45,6 @@ def test_dqn_test_expt(self):
             experiments, load_eval=False, exp_type="grid"
         )
 
-        import numpy as np
 
         final_metrics = np.squeeze(list_exp_data[0]["train_stats"])
         np.testing.assert_allclose(
@@ -56,13 +57,32 @@ def test_dqn_test_expt(self):
         exit_code = os.system("rm dqn_test_expt_0*.csv")
         assert exit_code == 0
 
-    def test_default_config(self):
+    # Similar thing is tested above. These tests are time consuming, so rather have only a few of them.
+    # def test_default_config(self):
 
-        exit_code = os.system(
-            sys.executable
-            + " mdp_playground/scripts/run_experiments.py -a 0 -n 0 -c default_config.py -e default_config"
-        )
-        assert exit_code == 0
+    #     exit_code = os.system(
+    #         sys.executable
+    #         + " mdp_playground/scripts/run_experiments.py -n 0 -c default_config.py -e default_config"
+    #     )
+    #     assert exit_code == 0
+
+    # ###TODO Enable once branches are merged
+    # def test_10_random_expts(self):
+
+    #     from glob import glob
+    #     expt_list = glob("experiments/*.py")
+
+    #     # sel_expt_list = np.random.randint(0, len(expt_list), 10)
+    #     expt_list = np.random.permutation(expt_list)
+    #     for i in range(2):
+    #         conf_file = expt_list[i]
+    #         exp_name = conf_file.split('/')[-1].split('.')[0]
+
+    #         exit_code = os.system(
+    #             sys.executable
+    #             + " run_experiments.py -n 0 -c " + conf_file + " -e " + exp_name + " -t 2000"
+    #         )
+    #         assert exit_code == 0
 
 
 if __name__ == "__main__":