Update all analysis scripts for new json (#4)

analysis/hist.py

@@ -23,6 +23,7 @@
 import os
 from glob import glob
 
+
 # third-party libs
 import dill
 import numpy as np
@@ -33,13 +34,15 @@
 from ilurl.envs.base import TrafficLightQLEnv
 
 ROOT_DIR = os.environ['ILURL_HOME']
-EMISSION_DIR = f"{ROOT_DIR}/data/emissions/"
-CONFIG_DIR = ('4545', '5040', '5434', '6030')
+# EMISSION_DIR = f"{ROOT_DIR}/data/emissions/"
+EMISSION_DIR = f"{ROOT_DIR}/data/experiments/0x04/"
+# CONFIG_DIR = ('4545', '5040', '5434', '6030')
+CONFIG_DIR = ('6030',)
 
 if __name__ == '__main__':
 
     # this loop acumulates experiments
-    ext = '.9000.w.info.json'
+    ext = '.9000.l.info.json'
     states = defaultdict(list)
 
     for config_dir in CONFIG_DIR:
@@ -55,11 +58,22 @@
             env = TrafficLightQLEnv.load(f"{filename}.pickle")
 
             # observation spaces
-            observation_spaces_per_cycle = output['observation_spaces']
-            for observation_space in observation_spaces_per_cycle:
-                for i, values in enumerate(env.ql_params.split_space(observation_space)):
+            if 'cycle' not in output:
+                # 0x00, 0x01, 0x02, 0x03
+                # deprecate
+                observation_spaces = [output['observation_spaces']]
+            else:
+                observation_spaces = output['observation_spaces']
+
+            for observation_space in observation_spaces:
+                variables = \
+                   env.ql_params.split_space(observation_space)
+                for i, values in enumerate(variables):
+
+                    ()
                     label = env.ql_params.states_labels[i]
-                    states[label] += values
+                    states[label] += \
+                        [val for value in values for val in value]
 
     # plot building
     num_bins = 50
@@ -72,7 +86,6 @@
         # mean and standard deviation of the distribution
         mu = np.mean(values)
         sigma = np.std(values)
-
         # the histogram of the data
         values_normalized = [
             round((v - mu) / sigma, 2) for v in values

analysis/q.py

@@ -10,16 +10,14 @@
 
 ROOT_DIR = os.environ['ILURL_HOME']
 # EMISSION_DIR = f"{ROOT_DIR}/data/experiments/0x02/"
-EMISSION_DIR = f"{ROOT_DIR}/data/experiments/0x03/"
+EMISSION_DIR = f"{ROOT_DIR}/data/experiments/0x04/"
 # CONFIG_DIRS = ('4545', '5040', '5434', '6030')
 CONFIG_DIRS = ('6030',)
 
 
-filenames = ('intersection_20200222-0900161582362016.287294.Q.1-49',
-             'intersection_20200222-0900371582362037.617882.Q.1-49',
-             'intersection_20200222-0900411582362041.078803.Q.1-49',
-             'intersection_20200222-0900481582362048.594145.Q.1-49',
-             'intersection_20200222-0900511582362051.183386.Q.1-49')
+filenames = ('intersection_20200227-2043431582836223.796797.Q.1-99',
+             'intersection_20200227-2128041582838884.952684.Q.1-100',
+             'intersection_20200227-2128511582838931.2348.Q.1-100')
 
 
 def qdist(Q, Q1):

analysis/rewards.py

@@ -1,7 +1,7 @@
 """Analyses aggregate experiment files e.g info and envs"""
 __author__ = 'Guilherme Varela'
 __date__ = '2020-02-13'
-import pdb
+
 import os
 import argparse
 import json
@@ -144,13 +144,13 @@ def str2bool(v):
 
         color = 'tab:blue'
         ax1.set_ylabel('Avg. speed', color=color)
-        ax1.plot(vels, color=color)
+        ax1.plot(np.mean(vels, axis=1), color=color)
         ax1.tick_params(axis='y', labelcolor=color)
 
         color = 'tab:red'
         ax2 = ax1.twinx()
         ax2.set_ylabel('Avg. vehicles', color=color)
-        ax2.plot(vehs, color=color)
+        ax2.plot(np.mean(vehs, axis=1), color=color)
         ax2.tick_params(axis='y', labelcolor=color)
 
         plt.title(f'{phase_split}:speed and count\n({db_type}, n={num_dbs})')
@@ -164,7 +164,7 @@ def str2bool(v):
 
         ax1.set_ylabel('Avg. Reward per Cycle', color=color)
         ax2.tick_params(axis='y', labelcolor=color)
-        ax1.plot(rets, color=color)
+        ax1.plot(np.mean(rets, axis=1), color=color)
         plt.title(f'{phase_split}:avg. cycle return\n({db_type}, n={num_dbs})')
         plt.savefig(f'{files_dir}{phase_split}_{db_type}_rewards.png')
         plt.show()
@@ -177,8 +177,6 @@ def str2bool(v):
         ax1.set_xlabel(f'Cycles ({cycle_time} sec)')
         ax1.set_ylabel('ratio optimal action')
 
-        import pdb
-        pdb.set_trace()
         cumacts = np.cumsum(acts == optact, axis=0)
         weights = np.arange(1, len(acts) + 1)
 

analysis/scatter.py

@@ -16,6 +16,7 @@
 import os
 from glob import glob
 
+
 # third-party libs
 import dill
 import numpy as np
@@ -27,13 +28,13 @@
 
 ROOT_DIR = os.environ['ILURL_HOME']
 # EMISSION_DIR = f"{ROOT_DIR}/data/experiments/0x02/"
-EMISSION_DIR = f"{ROOT_DIR}/data/experiments/0x03/"
+EMISSION_DIR = f"{ROOT_DIR}/data/experiments/0x04/"
 # CONFIG_DIR = ('4545', '5040', '5434', '6030')
 CONFIG_DIR = ('6030',)
 
 if __name__ == '__main__':
     # this loop acumulates experiments
-    ext = '.450000.l.info.json'
+    ext = '.9000.l.info.json'
     phases = defaultdict(list)
     labels = []
     desired_velocity = None
@@ -66,11 +67,17 @@
                 category_counts = env.ql_params.category_counts
 
             # observation spaces
-            observation_spaces_per_cycle = output['observation_spaces']
-            for intersection_space in observation_spaces_per_cycle:
-                for phase_space in intersection_space:
-                    for i, phase in enumerate(phase_space):
-                        phases[i] += [phase]
+            if 'cycle' not in output:
+                # deprecate: 0x00, 0x01, 0x02, 0x03
+                observation_spaces = [output['observation_spaces']]
+            else:
+                observation_spaces = output['observation_spaces']
+
+            for observation_space in observation_spaces:
+                for intersection_space in observation_space:
+                    for phase_space in intersection_space:
+                        for i, phase in enumerate(phase_space):
+                            phases[i] += [phase]
 
         _, ax = plt.subplots()
         for i, label in enumerate(labels):

ilurl/core/experiment.py

@@ -18,7 +18,7 @@
 import tempfile
 import time
 from collections import defaultdict
-import pdb
+
 
 from tqdm import tqdm
 

ilurl/core/ql/choice.py

@@ -1,7 +1,7 @@
 """The module helps define the q learning dictionary"""
 __author__ = "Guilherme Varela"
 __date__ = "2019-07-25"
-import pdb
+
 
 from numpy import argmax, sqrt, log, isnan
 from numpy.random import choice, rand

ilurl/core/ql/reward.py

@@ -2,7 +2,7 @@
 
 __author__ = "Guilherme Varela"
 __date__ = "2019-07-26"
-import pdb
+
 import numpy as np
 
 REWARD_TYPES = ('fix', 'weighted_average', 'score', 'target_velocity')