-
Notifications
You must be signed in to change notification settings - Fork 612
/
diagnose_model.py
370 lines (331 loc) · 13.3 KB
/
diagnose_model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
import matplotlib.pyplot as plt
import numpy
import seaborn
import torch
import models
from self_play import MCTS, Node, SelfPlay
class DiagnoseModel:
"""
Tools to understand the learned model.
Args:
weights: weights for the model to diagnose.
config: configuration class instance related to the weights.
"""
def __init__(self, checkpoint, config):
self.config = config
# Initialize the network
self.model = models.MuZeroNetwork(self.config)
self.model.set_weights(checkpoint["weights"])
self.model.to(
torch.device("cuda" if torch.cuda.is_available() else "cpu")
) # on GPU if available since the DataParallel objects in MuZeroNetwork requires that
self.model.eval()
def get_virtual_trajectory_from_obs(
self, observation, horizon, plot=True, to_play=0
):
"""
MuZero plays a game but uses its model instead of using the environment.
We still do an MCTS at each step.
"""
trajectory_info = Trajectoryinfo("Virtual trajectory", self.config)
root, mcts_info = MCTS(self.config).run(
self.model, observation, self.config.action_space, to_play, True
)
trajectory_info.store_info(root, mcts_info, None, numpy.NaN)
virtual_to_play = to_play
for i in range(horizon):
action = SelfPlay.select_action(root, 0)
# Players play turn by turn
if virtual_to_play + 1 < len(self.config.players):
virtual_to_play = self.config.players[virtual_to_play + 1]
else:
virtual_to_play = self.config.players[0]
# Generate new root
value, reward, policy_logits, hidden_state = self.model.recurrent_inference(
root.hidden_state,
torch.tensor([[action]]).to(root.hidden_state.device),
)
value = models.support_to_scalar(value, self.config.support_size).item()
reward = models.support_to_scalar(reward, self.config.support_size).item()
root = Node(0)
root.expand(
self.config.action_space,
virtual_to_play,
reward,
policy_logits,
hidden_state,
)
root, mcts_info = MCTS(self.config).run(
self.model, None, self.config.action_space, virtual_to_play, True, root
)
trajectory_info.store_info(
root, mcts_info, action, reward, new_prior_root_value=value
)
if plot:
trajectory_info.plot_trajectory()
return trajectory_info
def compare_virtual_with_real_trajectories(
self, first_obs, game, horizon, plot=True
):
"""
First, MuZero plays a game but uses its model instead of using the environment.
Then, MuZero plays the optimal trajectory according precedent trajectory but performs it in the
real environment until arriving at an action impossible in the real environment.
It does an MCTS too, but doesn't take it into account.
All information during the two trajectories are recorded and displayed.
"""
virtual_trajectory_info = self.get_virtual_trajectory_from_obs(
first_obs, horizon, False
)
real_trajectory_info = Trajectoryinfo("Real trajectory", self.config)
trajectory_divergence_index = None
real_trajectory_end_reason = "Reached horizon"
# Illegal moves are masked at the root
root, mcts_info = MCTS(self.config).run(
self.model,
first_obs,
game.legal_actions(),
game.to_play(),
True,
)
self.plot_mcts(root, plot)
real_trajectory_info.store_info(root, mcts_info, None, numpy.NaN)
for i, action in enumerate(virtual_trajectory_info.action_history):
# Follow virtual trajectory until it reaches an illegal move in the real env
if action not in game.legal_actions():
break # Comment to keep playing after trajectory divergence
action = SelfPlay.select_action(root, 0)
if trajectory_divergence_index is None:
trajectory_divergence_index = i
real_trajectory_end_reason = f"Virtual trajectory reached an illegal move at timestep {trajectory_divergence_index}."
observation, reward, done = game.step(action)
root, mcts_info = MCTS(self.config).run(
self.model,
observation,
game.legal_actions(),
game.to_play(),
True,
)
real_trajectory_info.store_info(root, mcts_info, action, reward)
if done:
real_trajectory_end_reason = "Real trajectory reached Done"
break
if plot:
virtual_trajectory_info.plot_trajectory()
real_trajectory_info.plot_trajectory()
print(real_trajectory_end_reason)
return (
virtual_trajectory_info,
real_trajectory_info,
trajectory_divergence_index,
)
def close_all(self):
plt.close("all")
def plot_mcts(self, root, plot=True):
"""
Plot the MCTS, pdf file is saved in the current directory.
"""
try:
from graphviz import Digraph
except ModuleNotFoundError:
print("Please install graphviz to get the MCTS plot.")
return None
graph = Digraph(comment="MCTS", engine="neato")
graph.attr("graph", rankdir="LR", splines="true", overlap="false")
id = 0
def traverse(node, action, parent_id, best):
nonlocal id
node_id = id
graph.node(
str(node_id),
label=f"Action: {action}\nValue: {node.value():.2f}\nVisit count: {node.visit_count}\nPrior: {node.prior:.2f}\nReward: {node.reward:.2f}",
color="orange" if best else "black",
)
id += 1
if parent_id is not None:
graph.edge(str(parent_id), str(node_id), constraint="false")
if len(node.children) != 0:
best_visit_count = max(
[child.visit_count for child in node.children.values()]
)
else:
best_visit_count = False
for action, child in node.children.items():
if child.visit_count != 0:
traverse(
child,
action,
node_id,
True
if best_visit_count and child.visit_count == best_visit_count
else False,
)
traverse(root, None, None, True)
graph.node(str(0), color="red")
# print(graph.source)
graph.render("mcts", view=plot, cleanup=True, format="pdf")
return graph
class Trajectoryinfo:
"""
Store the information about a trajectory (rewards, search information for every step, ...).
"""
def __init__(self, title, config):
self.title = title + ": "
self.config = config
self.action_history = []
self.reward_history = []
self.prior_policies = []
self.policies_after_planning = []
# Not implemented, need to store them in every nodes of the mcts
self.prior_values = []
self.values_after_planning = [[numpy.NaN] * len(self.config.action_space)]
self.prior_root_value = []
self.root_value_after_planning = []
self.prior_rewards = [[numpy.NaN] * len(self.config.action_space)]
self.mcts_depth = []
def store_info(self, root, mcts_info, action, reward, new_prior_root_value=None):
if action is not None:
self.action_history.append(action)
if reward is not None:
self.reward_history.append(reward)
self.prior_policies.append(
[
root.children[action].prior
if action in root.children.keys()
else numpy.NaN
for action in self.config.action_space
]
)
self.policies_after_planning.append(
[
root.children[action].visit_count / self.config.num_simulations
if action in root.children.keys()
else numpy.NaN
for action in self.config.action_space
]
)
self.values_after_planning.append(
[
root.children[action].value()
if action in root.children.keys()
else numpy.NaN
for action in self.config.action_space
]
)
self.prior_root_value.append(
mcts_info["root_predicted_value"]
if not new_prior_root_value
else new_prior_root_value
)
self.root_value_after_planning.append(root.value())
self.prior_rewards.append(
[
root.children[action].reward
if action in root.children.keys()
else numpy.NaN
for action in self.config.action_space
]
)
self.mcts_depth.append(mcts_info["max_tree_depth"])
def plot_trajectory(self):
name = "Prior policies"
print(name, self.prior_policies, "\n")
plt.figure(self.title + name)
ax = seaborn.heatmap(
self.prior_policies,
mask=numpy.isnan(self.prior_policies),
annot=True,
)
ax.set(xlabel="Action", ylabel="Timestep")
ax.set_title(name)
name = "Policies after planning"
print(name, self.policies_after_planning, "\n")
plt.figure(self.title + name)
ax = seaborn.heatmap(
self.policies_after_planning,
mask=numpy.isnan(self.policies_after_planning),
annot=True,
)
ax.set(xlabel="Action", ylabel="Timestep")
ax.set_title(name)
if 0 < len(self.action_history):
name = "Action history"
print(name, self.action_history, "\n")
plt.figure(self.title + name)
# ax = seaborn.lineplot(x=list(range(len(self.action_history))), y=self.action_history)
ax = seaborn.heatmap(
numpy.transpose([self.action_history]),
mask=numpy.isnan(numpy.transpose([self.action_history])),
xticklabels=False,
annot=True,
)
ax.set(ylabel="Timestep")
ax.set_title(name)
name = "Values after planning"
print(name, self.values_after_planning, "\n")
plt.figure(self.title + name)
ax = seaborn.heatmap(
self.values_after_planning,
mask=numpy.isnan(self.values_after_planning),
annot=True,
)
ax.set(xlabel="Action", ylabel="Timestep")
ax.set_title(name)
name = "Prior root value"
print(name, self.prior_root_value, "\n")
plt.figure(self.title + name)
# ax = seaborn.lineplot(x=list(range(len(self.prior_root_value))), y=self.prior_root_value)
ax = seaborn.heatmap(
numpy.transpose([self.prior_root_value]),
mask=numpy.isnan(numpy.transpose([self.prior_root_value])),
xticklabels=False,
annot=True,
)
ax.set(ylabel="Timestep")
ax.set_title(name)
name = "Root value after planning"
print(name, self.root_value_after_planning, "\n")
plt.figure(self.title + name)
# ax = seaborn.lineplot(x=list(range(len(self.root_value_after_planning))), y=self.root_value_after_planning)
ax = seaborn.heatmap(
numpy.transpose([self.root_value_after_planning]),
mask=numpy.isnan(numpy.transpose([self.root_value_after_planning])),
xticklabels=False,
annot=True,
)
ax.set(ylabel="Timestep")
ax.set_title(name)
name = "Prior rewards"
print(name, self.prior_rewards, "\n")
plt.figure(self.title + name)
ax = seaborn.heatmap(
self.prior_rewards, mask=numpy.isnan(self.prior_rewards), annot=True
)
ax.set(xlabel="Action", ylabel="Timestep")
ax.set_title(name)
if 0 < len(self.reward_history):
name = "Reward history"
print(name, self.reward_history, "\n")
plt.figure(self.title + name)
# ax = seaborn.lineplot(x=list(range(len(self.reward_history))), y=self.reward_history)
ax = seaborn.heatmap(
numpy.transpose([self.reward_history]),
mask=numpy.isnan(numpy.transpose([self.reward_history])),
xticklabels=False,
annot=True,
)
ax.set(ylabel="Timestep")
ax.set_title(name)
name = "MCTS depth"
print(name, self.mcts_depth, "\n")
plt.figure(self.title + name)
# ax = seaborn.lineplot(x=list(range(len(self.mcts_depth))), y=self.mcts_depth)
ax = seaborn.heatmap(
numpy.transpose([self.mcts_depth]),
mask=numpy.isnan(numpy.transpose([self.mcts_depth])),
xticklabels=False,
annot=True,
)
ax.set(ylabel="Timestep")
ax.set_title(name)
plt.show(block=False)