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
|
"""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)
from sdil.bci_metrics import (annotate_day_events, grouped_classification_accuracy,
grouped_regression_correlation, signature_metrics)
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")
def check_temporal_difference_role_vectorizer():
"""Role perturbations times performance change give the Harnett sign."""
cfg = BCIConfig(
days=2, episodes_per_day=4096, steps_per_episode=2,
feedback="performance_velocity", vectorizer_eta=0.5)
model = BCISDIL(cfg, model_seed=12)
generator = torch.Generator().manual_seed(13)
soma = 0.2 * torch.randn(
cfg.episodes_per_day, cfg.n_neurons, generator=generator)
xi = torch.empty_like(soma).bernoulli_(
0.5, generator=generator).mul_(2).sub_(1)
cursor_plus = (soma + cfg.perturb_sigma * xi) @ model.role
cursor_minus = (soma - cfg.perturb_sigma * xi) @ model.role
target = model.causal_role_targets(cursor_plus, cursor_minus, xi)
saved_role = model.role.clone()
model.role.zero_()
assert torch.equal(
target, model.causal_role_targets(cursor_plus, cursor_minus, xi))
model.role.copy_(saved_role)
mean_target = target.mean(0)
cosine = torch.nn.functional.cosine_similarity(
mean_target, model.role, dim=0).item()
assert cosine > 0.99
active = torch.ones(cfg.episodes_per_day, dtype=torch.bool)
model.A.zero_()
model.update_vectorizer(
torch.ones(cfg.episodes_per_day, 1),
torch.zeros(cfg.episodes_per_day, cfg.n_neurons),
target, active)
learned_cosine = torch.nn.functional.cosine_similarity(
model.A[:, 0], model.role, dim=0).item()
assert learned_cosine > 0.99
# With the neutral predictor exact, r_i = role_i * delta_error. Hence the
# P+ minus P- residual is positive in improving epochs and negative in
# worsening epochs, independent of instantaneous error magnitude.
model.A[:, 0].copy_(model.role)
model.P.copy_(model.coupling)
model.P_bias.zero_()
previous = torch.tensor([0.8, 0.4, 0.9, 0.3])
current = torch.tensor([0.6, 0.5, 0.5, 0.6])
delta = model.feedback_features(current, previous)
_, innovation, _, _ = model.apical_components(soma[:4], delta)
difference = (innovation[:, :cfg.n_plus].mean(1)
- innovation[:, cfg.n_plus:cfg.n_plus + cfg.n_minus].mean(1))
improving = delta[:, 0] > 0
worsening = delta[:, 0] < 0
sign_index = 0.5 * (
difference[improving].mean() - difference[worsening].mean())
assert sign_index > 0
print(
"BCI temporal-difference role vectorizer: "
f"cos={cosine:.6f}, sign_index={sign_index.item():.6f}")
def check_temporal_difference_plasticity_boundary():
cfg = BCIConfig(
days=2, episodes_per_day=32, steps_per_episode=6,
feedback="performance_velocity", forward_eta=0.03, kappa=0.0)
trajectories = generate_trajectories(cfg, 14)
initial = BCISDIL(cfg, model_seed=15)
initial.P.copy_(initial.coupling)
initial.A[:, 0].copy_(initial.role)
intact = initial.clone()
lesion = initial.clone()
report_intact = run_day(
intact, trajectories, 0, plasticity_gain=1.0,
control_gain=0.0, learn_vectorizer=False, learn_predictor=False,
collect=True)
report_control_sham = run_day(
initial.clone(), trajectories, 0, plasticity_gain=1.0,
control_gain=1.0, learn_vectorizer=False, learn_predictor=False,
collect=True)
run_day(
lesion, trajectories, 0, plasticity_gain=0.0,
control_gain=0.0, learn_vectorizer=False, learn_predictor=False)
assert not torch.equal(intact.W, initial.W)
assert torch.equal(lesion.W, initial.W)
for left, right in zip(
report_intact["events"], report_control_sham["events"]):
assert torch.equal(left["soma"], right["soma"])
print("BCI temporal-difference plasticity/online boundary: exact")
def check_grouped_decoders_and_metrics():
generator = torch.Generator().manual_seed(9)
groups = torch.arange(200)
x = torch.randn(200, 3, generator=generator)
labels = x[:, 0] + 0.2 * x[:, 1] > 0
accuracy, _ = grouped_classification_accuracy(x, labels, groups)
correlation = grouped_regression_correlation(x, 2 * x[:, 0] - x[:, 2], groups)
assert accuracy > 0.9
assert correlation > 0.99
cfg = BCIConfig(days=4, episodes_per_day=24, steps_per_episode=6,
kappa=0.1)
trajectories = generate_trajectories(cfg, 10)
model = BCISDIL(cfg, model_seed=11)
training_events = []
global_step = 0
for day in range(cfg.days):
report = run_day(
model, trajectories, day, collect=True, global_step=global_step)
global_step = report["global_step"]
annotate_day_events(
report["events"], day, report["success"],
episode_offset=day * cfg.episodes_per_day)
training_events.extend(report["events"])
evaluation = run_day(
model, trajectories, cfg.days - 1, collect=True,
global_step=global_step, plasticity_gain=0.0,
learn_vectorizer=False, learn_predictor=False)
annotate_day_events(evaluation["events"], 0, evaluation["success"], 10000)
metrics = signature_metrics(
training_events, evaluation["events"], cfg, model.role)
assert metrics["active_training_events"] > 0
assert metrics["evaluation_episodes"] == cfg.episodes_per_day
assert all(torch.isfinite(torch.tensor(value)) for value in metrics.values())
print(f"BCI grouped decoder mechanics: acc={accuracy:.3f}, corr={correlation:.3f}")
print("BCI preregistered signature metrics: finite")
if __name__ == "__main__":
check_roles_and_pairing()
check_causal_estimator()
check_predictor_identification()
check_phase_masks_and_velocity_reset()
check_temporal_difference_role_vectorizer()
check_temporal_difference_plasticity_boundary()
check_grouped_decoders_and_metrics()
print("ALL BCI MECHANICS CHECKS PASSED")
|
