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
|
"""CPU-only mechanics checks for the continuous Harnett-signature BCI."""
import os
import sys
import torch
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from sdil.bci import (BCIConfig, BCISDIL, causal_roles, generate_trajectories,
run_day)
def check_roles_and_pairing():
cfg = BCIConfig(days=2, episodes_per_day=8, steps_per_episode=4)
role = causal_roles(cfg)
assert torch.allclose(role[:5], torch.full((5,), 0.2))
assert torch.allclose(role[5:10], torch.full((5,), -0.2))
assert torch.equal(role[10:], torch.zeros(30))
first = generate_trajectories(cfg, 11)
second = generate_trajectories(cfg, 11)
third = generate_trajectories(cfg, 12)
assert torch.equal(first.context, second.context)
assert torch.equal(first.process_noise, second.process_noise)
assert torch.equal(first.perturbations, second.perturbations)
assert not torch.equal(first.context, third.context)
assert torch.equal(first.context[..., 0], torch.ones_like(first.context[..., 0]))
print("BCI causal roles and task-seed pairing: exact")
def check_causal_estimator():
cfg = BCIConfig(days=2, episodes_per_day=4096, steps_per_episode=2)
model = BCISDIL(cfg, model_seed=3)
generator = torch.Generator().manual_seed(4)
soma = torch.randn(4096, cfg.n_neurons, generator=generator) * 0.2
xi = torch.empty_like(soma).bernoulli_(0.5, generator=generator).mul_(2).sub_(1)
estimate = model.causal_targets(soma, cfg.target, xi)
exact = model.exact_causal_direction(soma, cfg.target)
mean_estimate = estimate.mean(0)
mean_exact = exact.mean(0)
cosine = torch.nn.functional.cosine_similarity(
mean_estimate, mean_exact, dim=0).item()
assert cosine > 0.99
assert estimate[:, 10:].abs().mean() > 0 # simultaneous cross-cell variance
assert mean_estimate[10:].abs().max() < 0.03
print(f"BCI simultaneous causal estimator mean cosine: {cosine:.6f}")
def check_predictor_identification():
cfg = BCIConfig(days=2, episodes_per_day=256, steps_per_episode=2,
predictor_eta=0.2)
model = BCISDIL(cfg, model_seed=5)
generator = torch.Generator().manual_seed(6)
active = torch.ones(256, dtype=torch.bool)
for _ in range(100):
soma = torch.randn(256, cfg.n_neurons, generator=generator).tanh()
ordinary = model.coupling * soma
model.update_predictor(soma, ordinary, active)
assert (model.P - model.coupling).abs().max() < 1e-4
assert model.P_bias.abs().max() < 1e-5
print("BCI neutral per-cell predictor identifies ordinary coupling: exact")
def check_phase_masks_and_velocity_reset():
cfg = BCIConfig(days=2, episodes_per_day=8, steps_per_episode=4,
kappa=0.3, feedback="error_velocity")
trajectories = generate_trajectories(cfg, 7)
initial = BCISDIL(cfg, model_seed=8)
intact = initial.clone()
online_lesion = initial.clone()
plasticity_lesion = initial.clone()
active = torch.ones(8, dtype=torch.bool)
args = (
trajectories.context[0, :, 0], trajectories.process_noise[0, :, 0],
trajectories.perturbations[0, :, 0], torch.zeros(8, cfg.n_neurons),
None, active, 1)
out_intact = intact.step(*args, control_gain=1.0, plasticity_gain=1.0,
learn_vectorizer=False)
out_online = online_lesion.step(*args, control_gain=0.0, plasticity_gain=1.0,
learn_vectorizer=False)
out_plastic = plasticity_lesion.step(
*args, control_gain=1.0, plasticity_gain=0.0, learn_vectorizer=False)
assert not torch.equal(out_intact["soma"], out_online["soma"])
assert torch.equal(intact.W, online_lesion.W)
assert torch.equal(out_intact["soma"], out_plastic["soma"])
assert torch.equal(plasticity_lesion.W, initial.W)
assert torch.equal(out_intact["feedback"][:, 1], torch.zeros(8))
day = run_day(initial.clone(), trajectories, 0, collect=True)
assert len(day["events"]) == cfg.steps_per_episode
assert day["global_step"] == cfg.steps_per_episode
print("BCI online/plasticity phase masks and episode velocity reset: exact")
if __name__ == "__main__":
check_roles_and_pairing()
check_causal_estimator()
check_predictor_identification()
check_phase_masks_and_velocity_reset()
print("ALL BCI MECHANICS CHECKS PASSED")
|
