Added timesteps_total to in run_experiments.py; some testing code

Author: RaghuSpaceRajan

misc/generate_mdpp_plots.py

@@ -0,0 +1,167 @@
+# Examples: 
+# py generate_mdpp_plots.py -f expt_list.txt
+# py generate_mdpp_plots.py --dir-name 13699485 --exp-name dqn_del # --show-plots  # dir_name and exp_name
+# Setup to analyse an MDP Playground experiment
+from mdp_playground.analysis import MDPP_Analysis
+
+import yaml
+import argparse
+
+from collections import Counter
+
+# Based on https://stackoverflow.com/a/71751051/11063709, to allow keys to have a list of values
+# in case duplicate keys are present in the YAML.
+def parse_preserving_duplicates(src):
+    class PreserveDuplicatesLoader(yaml.loader.Loader):
+        pass
+
+    def map_constructor(loader, node, deep=False):
+        keys = [loader.construct_object(node, deep=deep) for node, _ in node.value]
+        vals = [loader.construct_object(node, deep=deep) for _, node in node.value]
+        key_count = Counter(keys)
+        data = {}
+        for key, val in zip(keys, vals):
+            if key_count[key] > 1:
+                if key not in data:
+                    data[key] = []
+                data[key].append(val)
+            else:
+                data[key] = [val]
+        return data
+
+    PreserveDuplicatesLoader.add_constructor(yaml.resolver.BaseResolver.DEFAULT_MAPPING_TAG, map_constructor)
+    return yaml.load(src, PreserveDuplicatesLoader)
+
+
+def generate_plots(exp_name, dir_name, show_plots=False, options=''):
+    print("Generating plots for " + str(dir_name) + ": " + exp_name + " with the following addnl. options: " + options)
+    
+    # Set dir_name to the location where the CSV files from running an experiment were saved
+    dir_name = str(dir_name) # e.g. 13699485
+    # Set exp_name to the name that was given to the experiment when running it
+    # exp_name = 'dqn_del'
+    # Set the following to True to show plots that you generate below
+    # show_plots = True
+    # Set the following to True to save PDFs of plots that you generate below
+    save_fig = True
+    err_bar = 'bootstrap'  # 't_dist', 'std'
+    bonferroni = True
+    if 'normalise_episodic_reward' in options:
+        normalise_episodic_reward = True
+    else:
+        normalise_episodic_reward = False
+    if 'eval' in options:
+        load_eval = True
+    else:
+        load_eval = False
+    if 'auto_y_scale' in options:
+        common_y_scale = False
+    else:
+        common_y_scale = True
+
+    # Data loading
+    mdpp_analysis = MDPP_Analysis()
+    train_stats, eval_stats, train_curves, eval_curves, train_aucs, eval_aucs = mdpp_analysis.load_data(dir_name, exp_name, load_eval=load_eval, normalise_episodic_reward=normalise_episodic_reward)
+
+    # 1-D: Plots showing reward after total timesteps when varying a single meta-feature
+    # Plots across n runs: Training: with std dev across the runs
+    mdpp_analysis.plot_1d_dimensions(train_aucs, save_fig, bonferroni=bonferroni, err_bar=err_bar, show_plots=show_plots, common_y_scale=common_y_scale)
+
+    if 'ep_len' in options:
+        mdpp_analysis.plot_1d_dimensions(train_aucs, save_fig, bonferroni=bonferroni, err_bar=err_bar, show_plots=show_plots, metric_num=-1)
+
+    # 2-D heatmap plots across n runs: Training runs: with std dev across the runs
+    # There seems to be a bug with matplotlib - x and y axes tick labels are not correctly set even though we pass them. Please feel free to look into the code and suggest a correction if you find it.
+    if 'plot_2d' in options:
+        mdpp_analysis.plot_2d_heatmap(train_aucs, save_fig, show_plots=show_plots, common_y_scale=common_y_scale)
+
+        if 'ep_len' in options:
+            mdpp_analysis.plot_2d_heatmap(train_aucs, save_fig, show_plots=show_plots, common_y_scale=common_y_scale, metric_num=-1)
+
+    # Plot learning curves: Training: Each curve corresponds to a different seed for the agent
+    if 'learn_curves' in options:
+        mdpp_analysis.plot_learning_curves(train_curves, save_fig, show_plots=show_plots, common_y_scale=common_y_scale)
+
+    if 'eval' in options:
+        mdpp_analysis.plot_1d_dimensions(eval_aucs, save_fig, bonferroni=bonferroni, err_bar=err_bar, show_plots=show_plots, common_y_scale=common_y_scale, train=False)
+
+        if 'ep_len' in options:
+            mdpp_analysis.plot_1d_dimensions(eval_aucs, save_fig, bonferroni=bonferroni, err_bar=err_bar, show_plots=show_plots, metric_num=-1, train=False)
+
+        if 'plot_2d' in options:
+            mdpp_analysis.plot_2d_heatmap(eval_aucs, save_fig, show_plots=show_plots, common_y_scale=common_y_scale, train=False)
+
+            if 'ep_len' in options:
+                mdpp_analysis.plot_2d_heatmap(eval_aucs, save_fig, show_plots=show_plots, common_y_scale=common_y_scale, metric_num=-1, train=False)
+
+        # Plot learning curves: Training: Each curve corresponds to a different seed for the agent
+        if 'learn_curves' in options:
+            mdpp_analysis.plot_learning_curves(eval_curves, save_fig, show_plots=show_plots, common_y_scale=common_y_scale, train=False)
+
+
+if __name__ == "__main__":
+
+
+    parser = argparse.ArgumentParser(description="Process Latex .bib files")
+
+    parser.add_argument(
+        "--exp-file", "-f", type=str, help="Expt. identifiers and names listed in a YAML file, i.e., dir_name: exp_name",
+    )
+
+    parser.add_argument(
+        "--dir-name", "-d", type=str, help="dir name where expt. CSVs are stored"
+    )
+
+    parser.add_argument(
+        "--exp-name", "-e", type=str, help="expt name, corresponds to the names of the CSV stats files and the <config>.py file used for the expt."
+    )
+
+    parser.add_argument(
+        "--show-plots", "-p", action='store_true', dest='show_plots', help="Toggle displaying plots", default=False,
+    )
+
+    parser.add_argument(
+        "--num-expts", "-n", type=int, help="First n expts in the list are plotted"
+    )
+
+    args = parser.parse_args()
+
+    # print(args)
+
+    if args.exp_file is not None:
+        with open(args.exp_file) as f:
+            yaml_dict = parse_preserving_duplicates(f)  # yaml.safe_load(f)
+
+        print("List of expts.:", yaml_dict)
+
+        i = 0
+        for dir_name in yaml_dict:
+            if len(yaml_dict[dir_name]) > 1:
+                print("More than 1 expt. for the same expt_id:", dir_name, ". The expts.:", yaml_dict[dir_name])
+            for j in range(len(yaml_dict[dir_name])):
+                i += 1
+                print("\nExpt. no.:", i , "from the list.")
+
+                exp_name = yaml_dict[dir_name][j].split(' ')[0]
+                options = ' '.join(yaml_dict[dir_name][j].split(' ')[1:]) if ' ' in yaml_dict[dir_name][j] else ''
+                # if 'learn_curves' in options:
+                # if 'breakout_r' in exp_name:
+                generate_plots(dir_name=dir_name, exp_name=exp_name, show_plots=args.show_plots, options=options)
+
+                # Need to break out of 2 for loops
+                if args.num_expts is not None and i == args.num_expts:
+                    break
+
+            if args.num_expts is not None and i == args.num_expts:
+                break
+
+
+    else:
+        dict_args = vars(args)
+        del dict_args['exp_file']
+        dict_args['dir_name'] = dict_args['dir_name'].split(' ')[0]
+        dict_args['options'] = ' '.join(dict_args['dir_name'].split(' ')[1:]) if ' ' in dict_args['dir_name'] else ''
+        # print(dict_args)
+        generate_plots(**dict_args)
+
+

misc/test_expt_list.txt

@@ -0,0 +1,3 @@
+/home/rajanr/mdpp_15786014: dqn_seq_del learn_curves normalise_episodic_reward plot_2d auto_y_scale
+/home/rajanr/mdpp_15786214: rainbow_seq_del normalise_episodic_reward plot_2d auto_y_scale
+/home/rajanr/mdpp_15786316: a3c_seq_del normalise_episodic_reward plot_2d auto_y_scale