File size: 15,737 Bytes
cfbe945
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
"""Tabulate prescribed channels and source/receiver position contrasts."""
from common import CFG, HERE, OUT, config_digest, csv_write, dump, response
import json
import numpy as np


def groups(source_layout, receiver_layout, layout_config):
    sources = layout_config["original_source_centres"] if source_layout == 0 else CFG["source_centres"]
    receivers = layout_config["original_receiver_centres"] if receiver_layout == 0 else CFG["receiver_centres"]
    same = np.asarray([[(r[0] < .5) == (s[0] < .5) for s in sources] for r in receivers])
    answer = [("aggregate", 0, 0, np.ones((4, 3), dtype=bool)),
              ("same_side", 0, 0, same), ("opposite_sides", 0, 0, ~same)]
    for s in range(3):
        for r in range(4):
            mask = np.zeros((4, 3), dtype=bool)
            mask[r, s] = True
            answer.append(("channel", s + 1, r + 1, mask))
    return answer


def summarize():
    layout_config = json.loads((HERE / "layout_config.json").read_text())
    reference = np.load(OUT / "factorial_reference.npz")
    times = reference["time"]
    reference = reference["receiver_pressure"]
    accuracy, budgets, pairs = [], [], []
    for n in CFG["grids"]:
        full = np.load(OUT / f"n{n}_full.npz")
        field = np.load(OUT / f"n{n}_pressure_errors.npz")
        for a in range(2):
            for b in range(2):
                yr, yf = full["factorial_receiver_pressure"][a, b], reference[a, b]
                for end in CFG["window_ends"]:
                    active = times <= end + 1e-12
                    for group, s, r, channels in groups(a, b, layout_config):
                        ref = yf[active][:, channels]
                        err = yr[active][:, channels] - ref
                        accuracy.append({"grid": n, "method": "full", "modes_per_patch": 0, "dimension": n * n,
                            "source_layout": a, "receiver_layout": b, "time_end": end,
                            "group": group, "source": s, "receiver": r,
                            "total_receiver_error": float(np.linalg.norm(err) / np.linalg.norm(ref)),
                            "same_grid_receiver_error": 0., "reference_squared_norm": float(np.sum(ref ** 2)),
                            "continuous_pressure_error": float(np.sqrt(field["full_continuous_error_squared"][active].sum() /
                                field["continuous_norm_squared"][active].sum())) if a == 1 and group == "aggregate" else ""})
        for m in CFG["modes_per_patch"]:
            model = np.load(OUT / f"n{n}_cb_m{m}.npz")
            wr, wh, wf = (model[key] for key in ("matched_reduced_omega", "matched_full_omega", "matched_continuous_omega"))
            for j, (ri, hi, fi) in enumerate(zip(model["matched_reduced_indices"], model["matched_full_indices"], model["matched_continuous_indices"])):
                pairs.append({"grid": n, "modes_per_patch": m, "dimension": len(model["omega"]),
                    "reduced_index": int(ri), "full_index": int(hi), "continuous_index": int(fi),
                    "mac_reduced_full": float(model["mac_reduced_full"][j]),
                    "mac_full_continuous": float(model["mac_full_continuous"][j]),
                    "omega_reduced": float(wr[j]), "omega_full": float(wh[j]), "omega_continuous": float(wf[j]),
                    "relative_ritz_shift": float((wr[j] - wh[j]) / wf[j]),
                    "relative_grid_shift": float((wh[j] - wf[j]) / wf[j])})
            ritz_phase = np.cos(times[:, None] * wr[None]) - np.cos(times[:, None] * wh[None])
            grid_phase = np.cos(times[:, None] * wh[None]) - np.cos(times[:, None] * wf[None])
            for a in range(2):
                for b in range(2):
                    yf, yh = reference[a, b], full["factorial_receiver_pressure"][a, b]
                    yr = model["factorial_receiver_pressure"][a, b]
                    A, B = model["factorial_ritz_frequency"][a, b], model["factorial_grid_frequency"][a, b]
                    source, receiver, unmatched = (model["factorial_" + key][a, b] for key in ("source_coupling", "receiver_coupling", "unmatched"))
                    q = source + receiver + unmatched
                    Hr = model["factorial_reduced_residues"][a, b]
                    Hf = model["factorial_continuous_residues"][a, b]
                    omitted = yf - response(times, wf, Hf)
                    for end in CFG["window_ends"]:
                        active = times <= end + 1e-12
                        single_mode = np.sum(-ritz_phase[active] ** 2 - 2 * ritz_phase[active] * grid_phase[active], axis=0)
                        for group, s, r, channels in groups(a, b, layout_config):
                            ref = yf[active][:, channels]
                            er = (yr - yf)[active][:, channels]
                            er0 = (yr - A - yf)[active][:, channels]
                            er_cont = (yr - A - B - yf)[active][:, channels]
                            aa, bb, qq = (item[active][:, channels] for item in (A, B, q))
                            den = float(np.sum(ref ** 2))
                            exact = float((np.sum(er0 ** 2) - np.sum(er ** 2)) / den)
                            diagonal = float(single_mode @ np.sum(Hr[:, channels] ** 2, axis=1) / den)
                            frequency = float((-np.sum(aa ** 2) - 2 * np.sum(aa * bb)) / den)
                            cross_mode = frequency - diagonal
                            remainder_cross = float(-2 * np.sum(aa * qq) / den)
                            total = float(np.linalg.norm(er) / np.sqrt(den))
                            replaced = float(np.linalg.norm(er0) / np.sqrt(den))
                            key = {"grid": n, "method": "craig_bampton", "modes_per_patch": m, "dimension": len(model["omega"]),
                                   "source_layout": a, "receiver_layout": b, "time_end": end,
                                   "group": group, "source": s, "receiver": r}
                            same_grid = float(np.linalg.norm((yr - yh)[active][:, channels]) / np.linalg.norm(yh[active][:, channels]))
                            accuracy.append(key | {"total_receiver_error": total, "same_grid_receiver_error": same_grid,
                                "reference_squared_norm": den,
                                "continuous_pressure_error": float(np.sqrt(field[f"m{m}_continuous_error_squared"][active].sum() /
                                    field["continuous_norm_squared"][active].sum())) if a == 1 and group == "aggregate" else ""})
                            weights = np.linalg.norm(Hf[:, channels], axis=1)
                            weights /= max(weights.sum(), 1e-300)
                            budgets.append(key | {"accepted_triples": len(wr), "total_receiver_error": total,
                                "same_grid_receiver_error": same_grid, "coarse_replacement_error": replaced,
                                "continuous_replacement_error": float(np.linalg.norm(er_cont) / np.sqrt(den)),
                                "change_percentage_points": 100 * (replaced - total), "delta_squared_error": exact,
                                "same_mode_term": diagonal, "cross_mode_term": cross_mode, "remainder_cross_term": remainder_cross,
                                "source_cross_term": float(-2 * np.sum(aa * source[active][:, channels]) / den),
                                "receiver_cross_term": float(-2 * np.sum(aa * receiver[active][:, channels]) / den),
                                "unmatched_cross_term": float(-2 * np.sum(aa * unmatched[active][:, channels]) / den),
                                "budget_absolute_defect": abs(exact - diagonal - cross_mode - remainder_cross),
                                "ritz_relative_norm": float(np.linalg.norm(aa) / np.sqrt(den)),
                                "grid_relative_norm": float(np.linalg.norm(bb) / np.sqrt(den)),
                                "remainder_relative_norm": float(np.linalg.norm(qq) / np.sqrt(den)),
                                "matched_reference_omission": float(np.linalg.norm(omitted[active][:, channels]) / np.sqrt(den)),
                                "weighted_relative_grid_shift": float(weights @ ((wh - wf) / wf)),
                                "weighted_relative_ritz_shift": float(weights @ ((wr - wh) / wf)),
                                "reference_squared_norm": den})
    csv_write(OUT / "accuracy.csv", accuracy)
    csv_write(OUT / "finite_time_budgets.csv", budgets)
    csv_write(OUT / "modal_pairs.csv", pairs)
    primary = [row for row in budgets if row["source_layout"] == row["receiver_layout"] == 1]
    csv_write(OUT / "second_layout_budgets.csv", primary)
    cases = []
    for n in CFG["grids"]:
        for m in CFG["modes_per_patch"]:
            for end in CFG["window_ends"]:
                selected = [row for row in primary if row["grid"] == n and row["modes_per_patch"] == m and row["time_end"] == end]
                agg = next(row for row in selected if row["group"] == "aggregate")
                channels = [row for row in selected if row["group"] == "channel"]
                sides = {row["group"]: row["change_percentage_points"] for row in selected if row["group"] in ("same_side", "opposite_sides")}
                cases.append({"grid": n, "modes_per_patch": m, "time_end": end,
                    "total_receiver_error": agg["total_receiver_error"], "coarse_replacement_error": agg["coarse_replacement_error"],
                    "change_percentage_points": agg["change_percentage_points"], "accepted_triples": agg["accepted_triples"],
                    "beneficial_ritz_channels": sum(row["change_percentage_points"] > 1e-10 for row in channels),
                    "harmful_ritz_channels": sum(row["change_percentage_points"] < -1e-10 for row in channels),
                    "same_side_change_percentage_points": sides["same_side"],
                    "opposite_sides_change_percentage_points": sides["opposite_sides"]})
    summary = {"config_sha256": config_digest(), "accuracy_rows": len(accuracy), "budget_rows": len(budgets),
        "matched_mode_rows": len(pairs), "maximum_budget_absolute_defect": max(row["budget_absolute_defect"] for row in budgets),
        "second_layout_cases": cases,
        "factorial_T8_aggregate": [row for row in budgets if row["group"] == "aggregate" and row["time_end"] == 8.]}
    contrasts, example = position_contrasts(budgets)
    csv_write(OUT / "position_contrasts.csv", contrasts)
    summary["position_contrast_rows"] = len(contrasts)
    summary["channel_position_sign_reversals"] = sum(row["sign_reversal"] for row in contrasts if row["group"] == "channel")
    summary["aggregate_position_sign_reversals"] = [row for row in contrasts if row["group"] == "aggregate" and row["sign_reversal"]]
    summary["illustrative_position_reversal"] = example
    summary["figure_position_examples"] = figure_examples(budgets)
    dump(OUT / "summary.json", summary)
    return accuracy, budgets, summary


def position_contrasts(budgets):
    cases = [("source", (0, 0), (1, 0)), ("source", (0, 1), (1, 1)),
             ("receiver", (0, 0), (0, 1)), ("receiver", (1, 0), (1, 1))]
    index = {(row["grid"], row["modes_per_patch"], row["time_end"], row["source_layout"], row["receiver_layout"],
              row["group"], row["source"], row["receiver"]): row for row in budgets}
    rows = []
    for n in CFG["grids"]:
        for m in CFG["modes_per_patch"]:
            for end in CFG["window_ends"]:
                for changed, before, after in cases:
                    before_aggregate = index[(n, m, end, *before, "aggregate", 0, 0)]
                    after_aggregate = index[(n, m, end, *after, "aggregate", 0, 0)]
                    norm = np.sqrt(before_aggregate["reference_squared_norm"] * after_aggregate["reference_squared_norm"])
                    for group, s, r in [("aggregate", 0, 0)] + [("channel", s, r) for s in range(1, 4) for r in range(1, 5)]:
                        first, second = index[(n, m, end, *before, group, s, r)], index[(n, m, end, *after, group, s, r)]
                        d0, d1 = first["delta_squared_error"], second["delta_squared_error"]
                        reversal = d0 * d1 < 0 and min(abs(d0), abs(d1)) > 1e-12
                        strength = min(abs(d0 * first["reference_squared_norm"]), abs(d1 * second["reference_squared_norm"])) / norm
                        rows.append({"grid": n, "modes_per_patch": m, "time_end": end, "changed_positions": changed,
                            "before_source_layout": before[0], "before_receiver_layout": before[1],
                            "after_source_layout": after[0], "after_receiver_layout": after[1],
                            "group": group, "source": s, "receiver": r, "sign_reversal": int(reversal),
                            "before_change_percentage_points": first["change_percentage_points"],
                            "after_change_percentage_points": second["change_percentage_points"],
                            "before_delta_squared_error": d0, "after_delta_squared_error": d1,
                            "before_reference_energy_fraction": first["reference_squared_norm"] / before_aggregate["reference_squared_norm"],
                            "after_reference_energy_fraction": second["reference_squared_norm"] / after_aggregate["reference_squared_norm"],
                            "display_strength": float(strength)})
    candidates = [row for row in rows if row["group"] == "channel" and row["sign_reversal"]]
    example = max(candidates, key=lambda row: row["display_strength"]) if candidates else None
    if example is not None:
        example = example.copy()
        base = (example["grid"], example["modes_per_patch"], example["time_end"])
        example["before_budget"] = index[(*base, example["before_source_layout"], example["before_receiver_layout"],
            "channel", example["source"], example["receiver"])]
        example["after_budget"] = index[(*base, example["after_source_layout"], example["after_receiver_layout"],
            "channel", example["source"], example["receiver"])]
    return rows, example


def figure_examples(budgets):
    contrasts, _ = position_contrasts(budgets)
    selected = []
    for changed in ("source", "receiver"):
        candidates = [row for row in contrasts if row["group"] == "channel" and row["sign_reversal"]
            and row["changed_positions"] == changed]
        chosen = max(candidates, key=lambda row: row["display_strength"])
        first = next(row for row in budgets if row["group"] == "channel" and row["grid"] == chosen["grid"]
            and row["modes_per_patch"] == chosen["modes_per_patch"] and row["time_end"] == chosen["time_end"]
            and row["source"] == chosen["source"] and row["receiver"] == chosen["receiver"]
            and row["source_layout"] == chosen["before_source_layout"] and row["receiver_layout"] == chosen["before_receiver_layout"])
        second = next(row for row in budgets if row["group"] == "channel" and row["grid"] == chosen["grid"]
            and row["modes_per_patch"] == chosen["modes_per_patch"] and row["time_end"] == chosen["time_end"]
            and row["source"] == chosen["source"] and row["receiver"] == chosen["receiver"]
            and row["source_layout"] == chosen["after_source_layout"] and row["receiver_layout"] == chosen["after_receiver_layout"])
        selected.append(chosen | {"before_budget": first, "after_budget": second})
    return selected



def main():
    _, _, summary = summarize()
    print(json.dumps(summary, indent=2))


if __name__ == "__main__":
    main()