stead/plot_train_label_distribution.py

from __future__ import annotations

import argparse
import json
from pathlib import Path

import h5py
import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from matplotlib.ticker import MaxNLocator, ScalarFormatter


SCRIPT_DIR = Path(__file__).resolve().parent
DEFAULT_TRAIN_H5 = SCRIPT_DIR / "train" / "stead_100hz_60s_train.h5"
DEFAULT_TRAIN_META = SCRIPT_DIR / "train" / "stead_100hz_60s_train_meta.csv"
DEFAULT_GENERATION_META = SCRIPT_DIR / "train" / "stead_100hz_60s_train_meta_uniform10k.csv"
DEFAULT_STATION_CODES = SCRIPT_DIR / "station_codes_stead.csv"
DEFAULT_OUTPUT_PREFIX = SCRIPT_DIR / "stead_train_label_distribution_final_style"
DEFAULT_GENERATION_CSV = SCRIPT_DIR / "stead_generation_conditions_from_train.csv"

TRAIN_FILL = "#d8d6cf"
TRAIN_EDGE = "#5f5f5f"
GEN_EDGE = "#2f9cc8"
GRID_COLOR = "#e7e7e7"


def configure_matplotlib() -> None:
    mpl.rcParams.update(
        {
            "font.family": "sans-serif",
            "font.sans-serif": ["Arial", "Helvetica", "DejaVu Sans", "sans-serif"],
            "svg.fonttype": "none",
            "pdf.fonttype": 42,
            "font.size": 7,
            "axes.spines.right": False,
            "axes.spines.top": False,
            "axes.linewidth": 0.7,
            "axes.labelsize": 8,
            "axes.titlesize": 9,
            "xtick.labelsize": 7,
            "ytick.labelsize": 7,
            "legend.frameon": False,
            "xtick.major.width": 0.6,
            "ytick.major.width": 0.6,
            "xtick.major.size": 2.5,
            "ytick.major.size": 2.5,
        }
    )


def load_labels(train_h5: Path) -> np.ndarray:
    if not train_h5.exists():
        raise FileNotFoundError(f"Training HDF5 not found: {train_h5}")
    with h5py.File(train_h5, "r") as handle:
        if "labels" not in handle:
            raise KeyError(f"HDF5 file does not contain a 'labels' dataset: {train_h5}")
        labels = handle["labels"][:]
    if labels.ndim != 2 or labels.shape[1] < 4:
        raise ValueError(f"Expected labels with shape (n, >=4), got {labels.shape}")
    return labels.astype(np.float32, copy=False)


def load_labels_from_meta(train_meta: Path) -> np.ndarray:
    if not train_meta.exists():
        raise FileNotFoundError(f"Training metadata CSV not found: {train_meta}")
    columns = ["station_code", "p_index", "s_index", "s_decay"]
    frame = pd.read_csv(train_meta, usecols=columns)
    if frame.empty:
        raise ValueError(f"Training metadata CSV is empty: {train_meta}")
    for column in columns:
        values = pd.to_numeric(frame[column], errors="raise")
        if not np.isfinite(values).all():
            raise ValueError(f"Non-finite values found in metadata column: {column}")
        frame[column] = values
    return frame[columns].to_numpy(dtype=np.float32)


def draw_histogram(
    ax,
    train_values: np.ndarray,
    generation_values: np.ndarray,
    bins,
    title: str,
    xlabel: str,
    density: bool = True,
) -> None:
    ax.hist(
        train_values,
        bins=bins,
        density=density,
        histtype="stepfilled",
        color=TRAIN_FILL,
        edgecolor=TRAIN_EDGE,
        linewidth=0.9,
        alpha=0.72,
        label="Training",
    )
    ax.hist(
        generation_values,
        bins=bins,
        density=density,
        histtype="step",
        color=GEN_EDGE,
        linewidth=1.0,
        linestyle="--",
        label="Generation",
    )
    ax.grid(axis="y", color=GRID_COLOR, linewidth=0.45)
    ax.set_axisbelow(True)
    ax.set_title(title, loc="left", pad=3)
    ax.set_xlabel(xlabel)
    ax.set_ylabel("Density" if density else "Count")
    ax.xaxis.set_major_locator(MaxNLocator(nbins=4))
    ax.yaxis.set_major_locator(MaxNLocator(nbins=4))
    y_formatter = ScalarFormatter(useMathText=True)
    y_formatter.set_powerlimits((-2, 2))
    ax.yaxis.set_major_formatter(y_formatter)
    ax.yaxis.get_offset_text().set_fontsize(6)


def add_panel_label(ax, label: str) -> None:
    ax.text(
        -0.18,
        1.06,
        label,
        transform=ax.transAxes,
        fontsize=10,
        fontweight="bold",
        va="bottom",
        ha="left",
    )


def build_figure(train_labels: np.ndarray, generation_labels: np.ndarray):
    configure_matplotlib()
    fig, axes = plt.subplots(2, 3, figsize=(7.2, 3.7), constrained_layout=True)
    axes = axes.ravel()
    train_sp = train_labels[:, 2] - train_labels[:, 1]
    generation_sp = generation_labels[:, 2] - generation_labels[:, 1]
    train_decay_after_s = train_labels[:, 3] - train_sp
    generation_decay_after_s = generation_labels[:, 3] - generation_sp

    draw_histogram(
        axes[0],
        train_labels[:, 1],
        generation_labels[:, 1],
        bins=np.arange(350, 951, 50),
        title="P arrival",
        xlabel="Samples",
    )
    draw_histogram(
        axes[1],
        train_labels[:, 2],
        generation_labels[:, 2],
        bins=np.arange(150, 4851, 100),
        title="S arrival",
        xlabel="Samples",
    )
    draw_histogram(
        axes[2],
        train_sp,
        generation_sp,
        bins=np.arange(0, 4201, 100),
        title="S-P",
        xlabel="Samples",
    )
    draw_histogram(
        axes[3],
        train_labels[:, 3],
        generation_labels[:, 3],
        bins=np.arange(0, 5701, 125),
        title="Decay time",
        xlabel="Samples",
    )
    draw_histogram(
        axes[4],
        train_decay_after_s,
        generation_decay_after_s,
        bins=np.arange(-250, 5001, 125),
        title="Decay after S",
        xlabel="Samples",
    )
    draw_histogram(
        axes[5],
        train_labels[:, 0],
        generation_labels[:, 0],
        bins=np.arange(-0.5, max(train_labels[:, 0].max(), generation_labels[:, 0].max()) + 1.5, 25),
        title="Station label",
        xlabel="Station label",
        density=True,
    )

    for panel_label, ax in zip(("a", "b", "c", "d", "e", "f"), axes):
        add_panel_label(ax, panel_label)

    handles, labels_text = axes[0].get_legend_handles_labels()
    fig.legend(
        handles,
        labels_text,
        loc="upper right",
        bbox_to_anchor=(0.995, 1.06),
        ncol=2,
        handlelength=1.8,
        columnspacing=1.0,
    )
    fig.suptitle("STEAD label distributions", x=0.045, y=1.06, ha="left", fontsize=10, fontweight="bold")
    return fig


def write_generation_csv(labels: np.ndarray, generation_csv: Path, row_count: int, seed: int) -> Path:
    if row_count > 0 and row_count < len(labels):
        rng = np.random.default_rng(seed)
        selected = rng.choice(len(labels), size=row_count, replace=False)
        labels = labels[selected]
    frame = pd.DataFrame(
        {
            "station_label": labels[:, 0].astype(np.int64),
            "p_arrival": labels[:, 1],
            "s_arrival": labels[:, 2],
            "decay_time": labels[:, 3],
        }
    )
    frame.to_csv(generation_csv, index=False)
    return generation_csv


def summarize(train_labels: np.ndarray, generation_labels: np.ndarray) -> dict:
    summary = {
        "label_columns": ["station_label", "p_arrival", "s_arrival", "decay_time"],
        "train_sample_count": int(train_labels.shape[0]),
        "generation_sample_count": int(generation_labels.shape[0]),
        "train_unique_station_count": int(np.unique(train_labels[:, 0].astype(np.int64)).shape[0]),
        "generation_unique_station_count": int(np.unique(generation_labels[:, 0].astype(np.int64)).shape[0]),
        "train_raw_labels": {},
        "generation_raw_labels": {},
    }
    for prefix, label_array in (("train", train_labels), ("generation", generation_labels)):
        sp_values = label_array[:, 2] - label_array[:, 1]
        decay_after_s_values = label_array[:, 3] - sp_values
        for name, values in {
            "station_label": label_array[:, 0],
            "p_arrival": label_array[:, 1],
            "s_arrival": label_array[:, 2],
            "decay_time": label_array[:, 3],
            "s_minus_p": sp_values,
            "decay_after_s": decay_after_s_values,
        }.items():
            summary[f"{prefix}_raw_labels"][name] = describe(values)
    return summary


def describe(values: np.ndarray) -> dict:
    return {
        "count": int(values.size),
        "min": float(np.min(values)),
        "q05": float(np.quantile(values, 0.05)),
        "q25": float(np.quantile(values, 0.25)),
        "median": float(np.quantile(values, 0.50)),
        "q75": float(np.quantile(values, 0.75)),
        "q95": float(np.quantile(values, 0.95)),
        "max": float(np.max(values)),
        "mean": float(np.mean(values)),
        "std": float(np.std(values)),
    }


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description="Plot STEAD training label distributions and prepare a generation CSV.")
    parser.add_argument("--train-meta", type=Path, default=DEFAULT_TRAIN_META)
    parser.add_argument("--generation-meta", type=Path, default=DEFAULT_GENERATION_META)
    parser.add_argument("--train-h5", type=Path, default=DEFAULT_TRAIN_H5)
    parser.add_argument("--use-h5-labels", action="store_true")
    parser.add_argument("--station-codes", type=Path, default=DEFAULT_STATION_CODES)
    parser.add_argument("--output-prefix", type=Path, default=DEFAULT_OUTPUT_PREFIX)
    parser.add_argument("--generation-csv", type=Path, default=DEFAULT_GENERATION_CSV)
    parser.add_argument("--generation-rows", type=int, default=0)
    parser.add_argument("--seed", type=int, default=0)
    return parser.parse_args()


def main() -> None:
    args = parse_args()
    train_labels = load_labels(args.train_h5) if args.use_h5_labels else load_labels_from_meta(args.train_meta)
    generation_labels = load_labels_from_meta(args.generation_meta)
    train_station_labels = np.unique(train_labels[:, 0].astype(np.int64))

    if args.station_codes.exists():
        station_frame = pd.read_csv(args.station_codes)
        known_station_count = station_frame["station_code"].nunique()
    else:
        known_station_count = None

    fig = build_figure(train_labels, generation_labels)
    args.output_prefix.parent.mkdir(parents=True, exist_ok=True)
    fig.savefig(f"{args.output_prefix}.png", dpi=300, bbox_inches="tight")
    fig.savefig(f"{args.output_prefix}.pdf", bbox_inches="tight")
    fig.savefig(f"{args.output_prefix}.svg", bbox_inches="tight")
    plt.close(fig)

    generation_csv = write_generation_csv(generation_labels, args.generation_csv, args.generation_rows, args.seed)
    summary = summarize(train_labels, generation_labels)
    summary.update(
        {
            "train_h5": str(args.train_h5),
            "train_meta": str(args.train_meta),
            "generation_meta": str(args.generation_meta),
            "label_source": "h5_labels" if args.use_h5_labels else "metadata_csv",
            "known_station_count": None if known_station_count is None else int(known_station_count),
            "observed_station_count": int(train_station_labels.shape[0]),
            "png": f"{args.output_prefix}.png",
            "pdf": f"{args.output_prefix}.pdf",
            "svg": f"{args.output_prefix}.svg",
            "generation_csv": str(generation_csv),
            "generation_rows": int(args.generation_rows),
        }
    )
    summary_path = args.output_prefix.with_suffix(".summary.json")
    with open(summary_path, "w", encoding="utf-8") as fp:
        json.dump(summary, fp, ensure_ascii=False, indent=2)

    print(f"PNG: {args.output_prefix}.png")
    print(f"PDF: {args.output_prefix}.pdf")
    print(f"SVG: {args.output_prefix}.svg")
    print(f"Summary: {summary_path}")
    print(f"Generation CSV: {generation_csv}")


if __name__ == "__main__":
    main()