| """Fit driver-space parameters by group and record per-group runtime summaries.""" |
| import json |
| from pathlib import Path |
| import numpy as np |
| import pandas as pd |
| from _common import numeric, output_path, parser, read_csv, require, save_csv, preflight_outputs |
| from driver_space_model import drspace |
| from annotate_interactions import validate_parameters |
|
|
|
|
| def fit_groups(df, output, audit_path, width=2., length=5., percentile=0.5, minimum_samples=50000, |
| maximum_samples=50000, max_iter=50, workers=1, seed=42): |
| preflight_outputs(output, audit_path) |
| df = df.copy() |
| require(df, ["cangle", "round_v", "x_local", "y_local", "v_rel"]) |
| numeric(df, ["cangle"], integer=True) |
| numeric(df, ["round_v", "x_local", "y_local", "v_rel"]) |
| if not df.cangle.isin([0, 1]).all() or df.v_rel.le(0).any(): |
| raise ValueError("Require binary cangle and positive relative speed") |
| if (not np.isfinite([width, length, percentile]).all() or min(width, length) <= 0 or not 0 <= percentile <= 100 |
| or minimum_samples < 1 or maximum_samples < minimum_samples or workers < 1 or max_iter < 0): |
| raise ValueError("Invalid inference settings") |
| if Path(output).resolve() == Path(audit_path).resolve() or Path(audit_path).exists(): |
| raise FileExistsError("Choose distinct new parameter and audit files") |
| if Path(output).exists(): |
| raise FileExistsError(output) |
| records, results = [], [] |
| for (category, speed), group in df.groupby(["cangle", "round_v"], sort=True): |
| record = {"cangle": int(category), "round_v": float(speed), "available_samples": len(group), |
| "sample_group_ids": sorted(group.Sample_Group_ID.astype(str).unique().tolist()) if "Sample_Group_ID" in group else None} |
| if len(group) < minimum_samples: |
| record.update(status="insufficient_samples", used_samples=0) |
| records.append(record) |
| continue |
| samples = group[["x_local", "y_local", "v_rel"]].rename(columns={"x_local": "x", "y_local": "y", "v_rel": "v"}) |
| if len(samples) > maximum_samples: |
| samples = samples.sample(n=maximum_samples, random_state=seed) |
| model = drspace(samples, width=width, length=length, percentile=percentile) |
| limits, parameters, standard_errors, p_values = model.Inference(max_iter=max_iter, workers=workers) |
| status = getattr(model, "inference_status", "unknown") |
| optimizer_failures = getattr(model, "optimizer_failures", []) |
| accepted = (status in ["converged_rounded_parameters", "cycle_resolved_by_legacy_p_values"] |
| and not optimizer_failures and np.isfinite(parameters).all() and np.all(parameters > 0)) |
| record.update(status=status, accepted=bool(accepted), used_samples=len(samples), |
| iterations=int(getattr(model, "completed_iterations", 0)), |
| error=getattr(model, "inference_error", None), optimizer_failures=optimizer_failures, |
| bounds=np.asarray(limits).tolist(), parameters=np.asarray(parameters).tolist()) |
| |
| if accepted: |
| results.append({"cangle": int(category), "round_v": float(speed)}) |
| results[-1].update(dict(zip(["rx_plus", "rx_minus", "ry_plus", "ry_minus", "bx_plus", "bx_minus", "by_plus", "by_minus"], parameters))) |
| results[-1].update(sample_size=len(samples), Fit_Status=status, Fit_Method="author_legacy_hybrid") |
| records.append(record) |
| complete = bool(records) and len(results) == len(records) |
| audit = {"complete": complete, "method": "Alternating curvature-grid scale updates and constrained conditional maximum-likelihood shape updates", |
| "width_m": width, "length_m": length, "initial_percentile": percentile, |
| "minimum_samples": minimum_samples, "maximum_samples": maximum_samples, "max_iter": max_iter, "workers": workers, |
| "downsample_seed": seed, "groups": records, |
| "uncertainty_note": "Approximate inverse-Hessian p-values are used to select among recurring parameter vectors; the parameter CSV reports fitted values, sample size and fit status."} |
| |
| def clean(value): |
| if isinstance(value, dict): return {k: clean(v) for k, v in value.items()} |
| if isinstance(value, list): return [clean(v) for v in value] |
| return None if isinstance(value, float) and not np.isfinite(value) else value |
| with output_path(audit_path) as temporary: |
| Path(temporary).write_text(json.dumps(clean(audit), indent=2, allow_nan=False), encoding="utf-8") |
| if not complete: |
| raise ValueError("One or more inference groups failed or were undersized; audit saved; no parameter CSV published") |
| result = pd.DataFrame(results) |
| validate_parameters(result) |
| save_csv(result, output) |
| return result |
|
|
|
|
| def main(): |
| p = parser(__doc__) |
| p.add_argument("--input", required=True, help="Balanced fitting CSV generated by balance_pair_samples.py") |
| p.add_argument("--output", required=True, help="New complete parameter CSV, compatible with annotate_interactions.py") |
| p.add_argument("--audit", required=True, help="New JSON with per-group status and sample counts") |
| p.add_argument("--width-m", type=float, default=2., help="Driver-space model width setting in meters") |
| p.add_argument("--length-m", type=float, default=5., help="Driver-space model length setting in meters") |
| p.add_argument("--initial-percentile", type=float, default=0.5, help="Initial np.percentile value: 0.5 selects the 0.5th percentile") |
| p.add_argument("--minimum-samples", type=int, default=50000) |
| p.add_argument("--maximum-samples", type=int, default=50000) |
| p.add_argument("--max-iter", type=int, default=50) |
| p.add_argument("--workers", type=int, default=1, help="Number of inference workers; default is 1") |
| p.add_argument("--seed", type=int, default=42) |
| a = p.parse_args() |
| fit_groups(read_csv(a.input), a.output, a.audit, a.width_m, a.length_m, a.initial_percentile, |
| a.minimum_samples, a.maximum_samples, a.max_iter, a.workers, a.seed) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
