File size: 7,872 Bytes
aecbd27 | 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 | """Chain-test harness (preregistration, section 4) on synthetic data (no download)."""
import json
import os
from pathlib import Path
import numpy as np
import pytest
from flysim.validation import chains as ch
from flysim.validation.manifest import SET_A, SET_B
CONNECTOME = Path(os.environ.get(
"FLYSIM_CONNECTOME", Path(__file__).parents[2] / "data" / "connectome-v1"))
BODIES = {"DNp01": [1, 2], "MDN": [3, 4, 5, 6], "DNp09": [7, 8]}
SIDES = {"DNp01": ["L", "R"], "MDN": ["L", "R", "L", "R"], "DNp09": ["L", "R"]}
def test_keys_and_task_counts():
assert ch.stim_key("LC16+LPLC1+LPLC2", 100.0) == "LC16+LPLC1+LPLC2/100Hz"
assert ch.control_key(55, 3) == "control-55/03"
assert ch.file_stem("LC16+LPLC1+LPLC2/25Hz") == "LC16_LPLC1_LPLC2__25Hz"
t = ch.tasks_of("primary")
# 3 chains x (5 rates x 30 + 30 set B at 100 Hz) + 3 sizes x 20 sets x 10 trials
assert len(t) == 3 * (5 * 30 + 30) + 3 * 20 * 10
assert len(set(t)) == len(t)
assert ("LC4/100Hz", 59) in t and ("LC4/25Hz", 30) not in t
assert ("control-115/19", 9) in t and ("control-115/19", 10) not in t
assert len(ch.tasks_of("lc16-alt")) == 5 * 30 + 30 + 20 * 10
assert ch.rate_of_key("control-94/00") == 100.0
assert ch.rate_of_key("LC9/25Hz") == 25.0
def _cond(key, rates_hz, *, trials=SET_A, late=0.0, net=(1000, 2), seed=0):
"""Synthetic summary: Poisson counts per body at the given mean rate per readout."""
rng = np.random.default_rng(seed)
tr = list(trials)
counts = {r: rng.poisson(rates_hz.get(r, 0.0) * ch.STIM_S, (len(tr), len(b))).tolist()
for r, b in BODIES.items()}
return {"key": key, "trials": tr, "bodies": BODIES, "sides": SIDES, "counts": counts,
"first_ms": {r: [10.0 if any(c) else None for c in counts[r]] for r in BODIES},
"late_readout": [i < late * len(tr) for i in range(len(tr))],
"net_stim": [net[0]] * len(tr), "net_late": [net[1]] * len(tr),
"net_stim_nonstim": [net[0] // 2] * len(tr), "net_late_nonstim": [net[1]] * len(tr)}
def _data(stimuli, gain, control_hz=0.5, **kw):
"""gain: stimulus -> readout -> rate at 100 Hz (scaled linearly with input rate)."""
conds = {}
for s in stimuli:
for i, r in enumerate(ch.RATES):
key = ch.stim_key(s, r)
trials = [*SET_A, *SET_B] if r == ch.PRIMARY_RATE else SET_A
conds[key] = _cond(key, {ro: g * r / 100 for ro, g in gain[s].items()},
trials=trials, seed=i, **kw)
for k in range(ch.CONTROL_SETS):
key = ch.control_key(ch.SIZES[s], k)
conds[key] = _cond(key, dict.fromkeys(BODIES, control_hz), trials=ch.CONTROL_TRIALS,
seed=100 + k)
return ch.Trials({"conditions": conds})
GOOD = {"LC4": {"DNp01": 40}, "LC16": {"MDN": 30, "DNp09": 2},
"LC9": {"DNp09": 25, "DNp01": 1, "MDN": 1}}
def test_all_primary_chains_pass():
t = _data(["LC4", "LC16", "LC9"], GOOD)
att = ch.judge_all(t)
for s in ("LC4", "LC16", "LC9"):
assert len(att[s]) == 1 and att[s][0].passed, att[s][0].crits
assert [c.name for c in att["LC9"][0].crits] == [
"C1 activation", "C2 beats control", "C3 dose response", "C4 stability",
"C5 DNp09 > DNp01", "C5 DNp09 > MDN"]
assert [c.name for c in att["LC4"][0].crits][-1] == "C4 stability" # no C5 for LC4
md = ch.render(t, att)
assert "LC4 → DNp01 (escape): **PASS (primary)**" in md
def test_c1_fails_and_lc9_relative_reading_passes():
gain = {**GOOD, "LC9": {"DNp09": 1.5, "DNp01": 0.0, "MDN": 0.0}}
att = ch.judge_all(_data(["LC4", "LC16", "LC9"], gain, control_hz=0.0))
lc9 = att["LC9"]
assert [a.label for a in lc9] == ["primary", "alternative: relative reading"]
assert not lc9[0].passed and lc9[0].crits[0].name == "C1 activation"
assert not lc9[0].crits[0].passed
assert lc9[1].passed and lc9[1].crits[0].name == "C2 beats control"
def test_lc4_has_no_alternative():
gain = {**GOOD, "LC4": {"DNp01": 0.5}}
att = ch.judge_all(_data(["LC4", "LC16", "LC9"], gain))
assert len(att["LC4"]) == 1 and not att["LC4"][0].passed
def test_lc16_alternatives_in_order():
bad = {**GOOD, "LC16": {"MDN": 1.0, "DNp09": 3.0}}
t = _data(["LC4", "LC16", "LC9"], bad)
att = ch.judge_all(t)
assert [a.label for a in att["LC16"]] == ["primary",
"alternative step 1: LC16 + LPLC1 + LPLC2"]
assert att["LC16"][1].crits is None # not run yet
assert "NOT JUDGED" in ch.render(t, att)
# step 1 run, fails on specificity; it makes specificity worse than LC16 alone,
# so step 2 uses LC16 alone (and still fails C5 there)
t2 = _data(["LC4", "LC16", "LC9", "LC16+LPLC1+LPLC2"],
{**bad, "LC16+LPLC1+LPLC2": {"MDN": 5.0, "DNp09": 9.0}})
a2 = ch.judge_all(t2)["LC16"]
assert [a.label for a in a2][1:] == ["alternative step 1: LC16 + LPLC1 + LPLC2",
"alternative step 2: relative reading"]
assert a2[2].stimulus == "LC16" and a2[2].relative and not a2[2].passed
# step 1 passes: no step 2
t3 = _data(["LC4", "LC16", "LC9", "LC16+LPLC1+LPLC2"],
{**bad, "LC16+LPLC1+LPLC2": {"MDN": 30.0, "DNp09": 2.0}})
a3 = ch.judge_all(t3)["LC16"]
assert len(a3) == 2 and a3[1].passed
def test_stability_criterion():
t = _data(["LC4"], GOOD, late=0.1) # 10% of trials with late readout spikes
crit = {c.name: c for c in ch.judge_chain(t, ch.CHAINS[0], "LC4")}
assert not crit["C4 stability"].passed
t = _data(["LC4"], GOOD, net=(1000, 10)) # ratio (10/0.2)/(1000/0.3) = 0.015
crit = {c.name: c for c in ch.judge_chain(t, ch.CHAINS[0], "LC4")}
assert not crit["C4 stability"].passed
assert ch.Trials.stability(t, ch.stim_key("LC4", 100.0))[1] == pytest.approx(0.015)
def test_rates_per_side_and_set():
t = _data(["LC4"], GOOD)
k = ch.stim_key("LC4", 100.0)
assert len(t.rates(k, "MDN")) == 30 and len(t.rates(k, "MDN", None)) == 60
both = t.rates(k, "MDN")
left, right = t.rates(k, "MDN", side="L"), t.rates(k, "MDN", side="R")
assert np.allclose(both, (left + right) / 2)
def _neurons():
types = ["VP"] * 40 + ["LC4"] * 3 + ["VP"] * 5
side = ["R"] * 30 + ["L"] * 10 + ["R"] * 3 + ["R"] * 5
sc = ["visual_projection"] * 43 + ["ol_intrinsic"] * 5
sign = [1] * 25 + [-1] * 5 + [1] * 18
return ch.Neurons(np.arange(100, 148, dtype=np.int64), np.asarray(types, object),
np.asarray(side, object), np.asarray(sc, object),
np.asarray(sign, np.int8))
def test_control_pool_rule_and_draw(monkeypatch):
n = _neurons()
pool = n.control_pool()
# right side, sign +1, visual_projection, not LC4: bodies 100..124
assert pool.tolist() == list(range(100, 125))
monkeypatch.setattr(ch, "SIZES", {"LC4": 3, "X": 5})
d1, d2 = ch.draw_controls(n), ch.draw_controls(n)
assert d1 == d2 and d1["pool_size"] == 25
assert all(len(s) == 3 and len(set(s)) == 3 and set(s) <= set(pool.tolist())
for s in d1["sets"]["3"])
assert len(d1["sets"]["5"]) == ch.CONTROL_SETS
def test_committed_controls_shape():
d = ch.load_controls()
assert d["seed"] == 20261103 and d["pool_size"] == 3533
assert {k: (len(v), {len(s) for s in v}) for k, v in d["sets"].items()} == {
"55": (20, {55}), "94": (20, {94}), "115": (20, {115}), "251": (20, {251})}
@pytest.mark.skipif(not CONNECTOME.exists(), reason="connectome-v1 not fetched")
def test_committed_controls_match_the_draw_rule():
n = ch.Neurons.load(CONNECTOME)
assert ch.draw_controls(n) == json.loads(ch.CONTROLS_JSON.read_text())
for s, size in ch.SIZES.items():
assert len(ch.targets_of(ch.stim_key(s, 100.0), n, ch.load_controls())) == size
|