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
|
#!/usr/bin/env python3
"""Endpoint-free checks for the v2 cold-start recovery."""
from dataclasses import replace
import os
import sys
import torch
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from experiments.bci_v2_recovery_run import (
TARGETS,
build_recovery_config,
)
from sdil.bci_v2 import BCIV2
def copy_state(source, target):
for name in ("W", "A", "coupling", "P", "P_bias", "critic"):
setattr(target, name, getattr(source, name).clone())
def check_dense_velocity_cold_start():
recovered_cfg = replace(
build_recovery_config(),
days=2,
episodes_per_day=8,
steps_per_episode=2,
target=10.0,
inertia=0.0,
process_noise=0.0,
forward_eta=0.1,
perturb_every=99,
)
failed_cfg = replace(recovered_cfg, velocity_reward_scale=0.25)
recovered = BCIV2(recovered_cfg, model_seed=1)
recovered.P.copy_(recovered.coupling)
recovered.A.copy_(recovered.role)
recovered.critic.zero_()
failed = BCIV2(failed_cfg, model_seed=1)
copy_state(recovered, failed)
context = torch.zeros(8, recovered_cfg.context_dim)
context[:, 0] = 1.0
noise = torch.zeros(8, recovered_cfg.n_neurons)
noise[:, :recovered_cfg.n_plus] = torch.linspace(
0.1, 0.8, 8
).unsqueeze(1)
perturbations = torch.ones_like(noise)
recovered_initial_w = recovered.W.clone()
failed_initial_w = failed.W.clone()
recovered_state = recovered.initial_episode_state(8)
failed_state = failed.initial_episode_state(8)
_, recovered_record = recovered.step(
context,
noise,
perturbations,
recovered_state,
global_step=1,
final_step=False,
learn_role=False,
probe_role=False,
learn_predictor=False,
learn_critic=False,
critic_enabled=False,
)
_, failed_record = failed.step(
context,
noise,
perturbations,
failed_state,
global_step=1,
final_step=False,
learn_role=False,
probe_role=False,
learn_predictor=False,
learn_critic=False,
critic_enabled=False,
)
assert torch.allclose(
recovered_record["td_innovation"],
4.0 * failed_record["td_innovation"],
)
assert torch.allclose(
recovered.W - recovered_initial_w,
4.0 * (failed.W - failed_initial_w),
atol=1e-7,
)
print("recovery dense cold-start drive is exactly 4x failed v2")
def check_fixed_target_ladder():
assert TARGETS == (1.55, 1.60, 1.65, 1.70, 1.75, 1.80)
maxima = torch.tensor([1.58, 1.63, 1.68, 1.73, 1.78])
fractions = torch.tensor([
(maxima >= target).float().mean() for target in TARGETS
])
assert torch.all(fractions[:-1] >= fractions[1:])
assert bool((fractions > 0).any()) and bool((fractions < 1).any())
print("recovery target ladder is fixed, ordered, and monotone")
if __name__ == "__main__":
check_dense_velocity_cold_start()
check_fixed_target_ladder()
print("ALL V2 COLD-START RECOVERY MECHANICS CHECKS PASSED")
|
