added first working MAPPO implementation

studies/normalization_study.py

@@ -1,14 +1,13 @@
 from algorithms.utils import Checkpointer
 from pathlib import Path
 from algorithms.utils import load_yaml_file, add_env_props, instantiate_class, load_class
-from algorithms.marl import LoopSNAC, LoopIAC, LoopSEAC
+#from algorithms.marl import LoopSNAC, LoopIAC, LoopSEAC
 
 
-#study_root = Path(__file__).parent / 'curious_study'
-study_root = Path('/Users/romue/PycharmProjects/EDYS/algorithms/marl')
 
 for i in range(0, 5):
-    for name in ['example_config']:
+    for name in ['mappo']:#['seac', 'iac', 'snac']:
+        study_root = Path(__file__).parent / name
         cfg = load_yaml_file(study_root / f'{name}.yaml')
         add_env_props(cfg)
 
@@ -17,7 +16,7 @@ for i in range(0, 5):
         max_steps = cfg['algorithm']['max_steps']
         n_steps = cfg['algorithm']['n_steps']
 
-        checkpointer = Checkpointer(f'{name}#{i}', study_root, cfg, max_steps, 250)
+        checkpointer = Checkpointer(f'{name}#{i}', study_root, cfg, max_steps, 50)
 
         loop = load_class(cfg['method'])(cfg)
         df = loop.train_loop(checkpointer)

studies/playground_file.py

@@ -1,32 +1,22 @@
-import numpy as np
 import pandas as pd
 from pathlib import Path
 import matplotlib.pyplot as plt
 import seaborn as sns
 
-study_root = Path(__file__).parent / 'entropy_study'
-names_all = ['basic_gru', 'layernorm_gru', 'spectralnorm_gru', 'nonorm_gru']
-names_only_1 = ['L2OnlyAh_gru', 'L2OnlyChAh_gru', 'L2OnlyMix_gru', 'basic_gru']
-names_only_2 = ['L2NoCh_gru', 'L2NoAh_gru', 'nomix_gru', 'basic_gru']
 
-names = names_only_2
-#names = ['nonorm_gru']
-# /Users/romue/PycharmProjects/EDYS/studies/normalization_study/basic_gru#3
-csvs = []
-for name in ['basic_gru', 'nonorm_gru', 'spectralnorm_gru']:
-    for run in range(0, 1):
+dfs = []
+for name in ['l2snac', 'iac', 'snac', 'seac']:
+    for c in range(5):
         try:
-            df = pd.read_csv(study_root / f'{name}#{run}' / 'results.csv')
-            df = df[df.agent == 'sum']
-            df = df.groupby(['checkpoint', 'run']).mean().reset_index()
-            df['method'] = name
-            df['run_'] = run
-
-            df.reward = df.reward.rolling(15).mean()
-            csvs.append(df)
+            study_root = Path(__file__).parent / name / f'{name}#{c}'
+            df = pd.read_csv(study_root / 'results.csv', index_col=False)
+            df.reward = df.reward.rolling(100).mean()
+            df['method'] = name.upper()
+            dfs.append(df)
         except Exception as e:
-            print(f'skipped {run}\t {name}')
+            pass
 
-csvs = pd.concat(csvs).rename(columns={"checkpoint": "steps*2e3", "B": "c"})
-sns.lineplot(data=csvs, x='steps*2e3', y='reward', hue='method', palette='husl', ci='sd', linewidth=1.8)
-plt.savefig('entropy.png')
+df = pd.concat(dfs).reset_index()
+sns.lineplot(data=df, x='episode', y='reward', hue='method', palette='husl', ci='sd', linewidth=1.5)
+plt.savefig('study.png')
+print('saved image')

studies/viz_policy.py

@@ -3,19 +3,21 @@ from algorithms.marl import LoopSNAC, LoopIAC, LoopSEAC
 from pathlib import Path
 from algorithms.utils import load_yaml_file
 from tqdm import trange
-study = 'curious_study'
-study_root = Path(__file__).parent / study
+study = 'example_config#0'
+#study_root = Path(__file__).parent / study
+study_root = Path('/Users/romue/PycharmProjects/EDYS/algorithms/marl/')
 
 #['L2NoAh_gru', 'L2NoCh_gru', 'nomix_gru']:
 render = True
 eval_eps = 3
 for run in range(0, 5):
-    for name in ['basic_gru']:#['L2OnlyAh_gru', 'L2OnlyChAh_gru', 'L2OnlyMix_gru']: #['layernorm_gru', 'basic_gru', 'nonorm_gru', 'spectralnorm_gru']:
-        cfg = load_yaml_file(Path(__file__).parent / study / f'{name}.yaml')
-        p_root = Path(study_root / f'{name}#{run}')
+    for name in ['example_config']:#['L2OnlyAh_gru', 'L2OnlyChAh_gru', 'L2OnlyMix_gru']: #['layernorm_gru', 'basic_gru', 'nonorm_gru', 'spectralnorm_gru']:
+        cfg = load_yaml_file(study_root / study / 'config.yaml')
+        #p_root = Path(study_root / study / f'{name}#{run}')
         dfs = []
         for i in trange(500):
-            path = p_root / f'checkpoint_{i}'
+            path = study_root / study / f'checkpoint_{161}'
+            print(path)
 
             snac = LoopSEAC(cfg)
             snac.load_state_dict(path)