"""Reproduce the measured comparisons and the regression path from model_segments.csv.
Needs only Python 3 and numpy:  python3 reproduce.py
The physics (road-load) fit is described in the method page; this script covers the
trip averages, the same-speed comparison and regressions M0 to M5 with drive-clustered errors."""
import csv
import numpy as np

rows = list(csv.DictReader(open("model_segments.csv")))
EXCLUDE = {"32"}  # Bodo drive with a stale position, kept out of the main models
f = lambda r, k: float(r[k])
data = []
for r in rows:
    if r["drive"] in EXCLUDE or r["gps_stale"] == "True" or f(r, "max_gap_s") > 10 or not r["rho"] or not r["dh_car"] or not r["wet"]:
        continue
    d, dh = f(r, "dist_m"), f(r, "dh_car")
    if abs(dh / d) > 0.12:  # implausible grade from the car's altitude
        continue
    data.append(dict(drive=r["drive"], tow=float(r["label"] == "tow"), kwh100=f(r, "e_kwh") / d * 1e5, km=d / 1000, e=f(r, "e_kwh"),
                     v=f(r, "v_mean_kmh"), inv_v=100 / f(r, "v_mean_kmh"), vsq=f(r, "S3") / d / (100 / 3.6) ** 2,
                     ke=0.5 * (f(r, "v1") ** 2 - f(r, "v0") ** 2) / d / 100, gup=max(dh, 0) / d * 100, gdn=min(dh, 0) / d * 100,
                     temp_c=f(r, "car_temp"), warm=float(int(r["seg"]) < 3), w=f(r, "headwind_base"), wetx=f(r, "wet")))
print(f"{len(data)} one-kilometre stretches in the models")

# Article figure 1, trip averages: whole drives (drives.csv), total kWh over total km
drv = [d for d in csv.DictReader(open("drives.csv")) if d["excluded_from_main_models"] == "0"]
for lab in ("solo", "tow"):
    s = [d for d in drv if d["caravan"] == lab]
    print(f"trip average {lab:5s} {len(s):3d} drives  {100 * sum(float(d['kwh']) for d in s) / sum(float(d['km']) for d in s):5.1f} kWh/100 km")
# Article figure 2, same speed: every stretch at 70 to 90 km/h except the Bodo drive (wider than the model sample)
for lab in ("solo", "tow"):
    s = [r for r in rows if r["drive"] not in EXCLUDE and r["label"] == lab and 70 <= f(r, "v_mean_kmh") < 90]
    print(f"70-90 km/h  {lab:5s} {len(s):4d} km  {100 * sum(f(r, 'e_kwh') for r in s) / sum(f(r, 'dist_m') / 1000 for r in s):5.1f} kWh/100 km")

def ols_cluster(cols):
    X = np.array([[1.0] + [x[c] for c in cols] for x in data]); y = np.array([x["kwh100"] for x in data])
    b, *_ = np.linalg.lstsq(X, y, rcond=None); e = y - X @ b
    XtXi = np.linalg.inv(X.T @ X); groups = {}
    for i, x in enumerate(data): groups.setdefault(x["drive"], []).append(i)
    meat = sum(np.outer(X[g].T @ e[g], X[g].T @ e[g]) for g in groups.values())
    n, k, G = len(y), X.shape[1], len(groups)
    V = XtXi @ meat @ XtXi * G / (G - 1) * (n - 1) / (n - k)
    return b, np.sqrt(np.diag(V)), 1 - (e @ e) / ((y - y.mean()) @ (y - y.mean()))

PATH = [("M0 towing only", ["tow"]), ("M1 + speed", ["tow", "inv_v", "vsq", "ke"]), ("M2 + slope", ["tow", "inv_v", "vsq", "ke", "gup", "gdn"]),
        ("M3 + temperature, warm-up", ["tow", "inv_v", "vsq", "ke", "gup", "gdn", "temp_c", "warm"]),
        ("M4 + headwind", ["tow", "inv_v", "vsq", "ke", "gup", "gdn", "temp_c", "warm", "w"]),
        ("M5 + wet road (modeled)", ["tow", "inv_v", "vsq", "ke", "gup", "gdn", "temp_c", "warm", "w", "wetx"])]
for name, cols in PATH:
    b, se, r2 = ols_cluster(cols)
    print(f"{name:28s} towing +{b[1]:5.1f} kWh/100 km (cluster SE {se[1]:.1f})  R2 {r2:.2f}")
