scripts/plot_xrd_distributions.py

#!/usr/bin/env python3
"""Plot MP-20 XRD distribution figures from previously generated statistics."""

from __future__ import annotations

import argparse
from pathlib import Path

import numpy as np
import pandas as pd

import matplotlib

matplotlib.use("Agg")
import matplotlib.pyplot as plt


def finite_series(values: pd.Series) -> pd.Series:
    values = pd.to_numeric(values, errors="coerce")
    return values[np.isfinite(values)]


def describe(values: pd.Series, name: str) -> dict[str, float | int | str]:
    values = finite_series(values)
    qs = values.quantile([0.001, 0.01, 0.05, 0.5, 0.95, 0.99, 0.999])
    return {
        "metric": name,
        "count": int(values.size),
        "min": float(values.min()),
        "q001": float(qs.loc[0.001]),
        "q01": float(qs.loc[0.01]),
        "q05": float(qs.loc[0.05]),
        "median": float(qs.loc[0.5]),
        "mean": float(values.mean()),
        "q95": float(qs.loc[0.95]),
        "q99": float(qs.loc[0.99]),
        "q999": float(qs.loc[0.999]),
        "max": float(values.max()),
    }


def add_quantile_lines(ax: plt.Axes, values: pd.Series, *, color: str = "#334155") -> None:
    values = finite_series(values)
    for q, label, style in [
        (0.5, "median", "-"),
        (0.95, "q95", "--"),
        (0.99, "q99", ":"),
    ]:
        x = float(values.quantile(q))
        ax.axvline(x, color=color, linestyle=style, linewidth=1.4, alpha=0.85)
        ax.text(
            x,
            ax.get_ylim()[1] * 0.96,
            f" {label}={x:.3g}",
            rotation=90,
            va="top",
            ha="left",
            fontsize=8,
            color=color,
        )


def save_hist(
    values: pd.Series,
    path: Path,
    *,
    title: str,
    xlabel: str,
    bins: int | np.ndarray = 80,
    log_x: bool = False,
    xlim: tuple[float, float] | None = None,
    color: str = "#2563eb",
) -> None:
    values = finite_series(values)
    fig, ax = plt.subplots(figsize=(8.5, 5.2), dpi=180)
    ax.hist(values, bins=bins, color=color, alpha=0.82, edgecolor="white", linewidth=0.35)
    ax.set_title(title)
    ax.set_xlabel(xlabel)
    ax.set_ylabel("Number of materials")
    if log_x:
        ax.set_xscale("log")
    if xlim is not None:
        ax.set_xlim(*xlim)
    add_quantile_lines(ax, values)
    ax.grid(True, axis="y", alpha=0.25)
    fig.tight_layout()
    fig.savefig(path)
    plt.close(fig)


def save_joint(last_df: pd.DataFrame, path: Path) -> None:
    x = finite_series(last_df["last_peak_2theta"])
    y = finite_series(last_df["last_peak_intensity"])
    aligned = pd.concat([x, y], axis=1).dropna()

    fig, ax = plt.subplots(figsize=(8.2, 5.8), dpi=180)
    hb = ax.hexbin(
        aligned["last_peak_2theta"],
        aligned["last_peak_intensity"],
        gridsize=70,
        cmap="viridis",
        mincnt=1,
        bins="log",
    )
    ax.set_title("Last XRD Peak Position vs. Relative Intensity")
    ax.set_xlabel("Last peak 2theta (degree)")
    ax.set_ylabel("Last peak relative intensity (max peak = 100)")
    for xv in [150, 160, 170, 178]:
        ax.axvline(xv, color="#e11d48", linestyle="--", linewidth=0.8, alpha=0.55)
        ax.text(xv, ax.get_ylim()[1] * 0.98, f" {xv}", rotation=90, va="top", fontsize=8, color="#9f1239")
    for yv in [1, 5, 10]:
        ax.axhline(yv, color="#0f766e", linestyle=":", linewidth=0.9, alpha=0.65)
        ax.text(ax.get_xlim()[0], yv, f" {yv}", va="bottom", fontsize=8, color="#115e59")
    cbar = fig.colorbar(hb, ax=ax)
    cbar.set_label("log10(number of materials)")
    ax.grid(True, alpha=0.18)
    fig.tight_layout()
    fig.savefig(path)
    plt.close(fig)


def build_last_peak_intensity(peaks_csv: Path, out_csv: Path, chunksize: int) -> pd.DataFrame:
    if out_csv.exists():
        return pd.read_csv(out_csv)

    best: dict[tuple[str, int, str], tuple[float, float]] = {}
    usecols = ["split", "index", "material_id", "two_theta", "intensity"]
    for chunk in pd.read_csv(peaks_csv, usecols=usecols, chunksize=chunksize):
        chunk["index"] = pd.to_numeric(chunk["index"], errors="coerce").astype("Int64")
        chunk["two_theta"] = pd.to_numeric(chunk["two_theta"], errors="coerce")
        chunk["intensity"] = pd.to_numeric(chunk["intensity"], errors="coerce")
        chunk = chunk.dropna(subset=["index", "two_theta", "intensity"])
        idx = chunk.groupby(["split", "index", "material_id"], sort=False)["two_theta"].idxmax()
        sub = chunk.loc[idx, usecols]
        for row in sub.itertuples(index=False):
            key = (row.split, int(row.index), row.material_id)
            candidate = (float(row.two_theta), float(row.intensity))
            if key not in best or candidate[0] > best[key][0]:
                best[key] = candidate

    rows = [
        {
            "split": split,
            "index": index,
            "material_id": material_id,
            "last_peak_2theta_from_peaks": two_theta,
            "last_peak_intensity": intensity,
        }
        for (split, index, material_id), (two_theta, intensity) in best.items()
    ]
    df = pd.DataFrame(rows).sort_values(["split", "index", "material_id"])
    df.to_csv(out_csv, index=False)
    return df


def write_readme(path: Path) -> None:
    text = """# MP-20 XRD 分布图说明

本目录中的图用于把之前的表格统计变成更直观的分布图，方便向老师解释 `2theta_max`、最后峰强度以及 bin size 的选择依据。

## 输入数据

- `../mp20_xrd_material_stats.csv`
  - 每个材料一行。
  - 使用其中的 `last_peak_2theta` 和 `min_peak_gap`。
- `../mp20_xrd_peaks.csv`
  - 每个峰一行。
  - 仅用于派生每个材料最后一个峰的强度。
  - 派生结果已保存为 `mp20_xrd_last_peak_intensity.csv`，后续重新画图不需要再次扫描大峰表。

## 输出图

1. `last_peak_2theta_distribution.png`
   - 横轴：每个材料最后一个 XRD 峰的位置，单位为 degree 2theta。
   - 纵轴：材料数量。
   - 图中竖线标出 median、q95、q99。

2. `last_peak_intensity_distribution.png`
   - 横轴：每个材料最后一个峰的相对强度。
   - pymatgen 中最强峰通常归一化为 100，因此这里的强度可以理解为“最后峰相对于最强峰的百分比”。
   - 如果大量最后峰强度接近 0，说明高角度最后峰可能很弱。

3. `last_peak_2theta_vs_intensity.png`
   - 横轴：最后峰位置。
   - 纵轴：最后峰相对强度。
   - 使用 hexbin 显示二维密度，颜色越亮表示材料越多。
   - 红色竖线是常见 2theta 截断参考位置；绿色横线是强度阈值参考。
   - 这张图用于判断“大角度最后峰是否主要是弱峰”。

4. `min_peak_gap_distribution.png`
   - 横轴：每个材料内部相邻峰之间的最小间距，单位为 degree 2theta。
   - 纵轴：材料数量。
   - 因为最小 gap 有极小异常值，所以主图使用 log x 轴。
   - 这张图用于判断离散化 bin size 是否会被极端材料过度约束。

## 汇总表

`distribution_summary.csv` 保存四个变量的常用统计量：

- `count`
- `min`
- `q001`
- `q01`
- `q05`
- `median`
- `mean`
- `q95`
- `q99`
- `q999`
- `max`

## 注意

第 4 张图使用的是“每个材料的最小峰间距分布”，不是所有相邻峰 gap 的总体分布。这样更符合当前任务：每个材料贡献一个最小 gap，避免峰数很多的材料在总体分布中占过大权重。
"""
    path.write_text(text, encoding="utf-8")


def main() -> None:
    parser = argparse.ArgumentParser()
    parser.add_argument("--stats-dir", type=Path, default=Path("/workspace/mp-20/xrd_stats"))
    parser.add_argument("--output-dir", type=Path, default=Path("/workspace/mp-20/xrd_stats/distributions"))
    parser.add_argument("--chunksize", type=int, default=500_000)
    args = parser.parse_args()

    args.output_dir.mkdir(parents=True, exist_ok=True)
    material_csv = args.stats_dir / "mp20_xrd_material_stats.csv"
    peaks_csv = args.stats_dir / "mp20_xrd_peaks.csv"
    last_intensity_csv = args.output_dir / "mp20_xrd_last_peak_intensity.csv"

    material = pd.read_csv(material_csv)
    material = material[material["status"].eq("ok")].copy()
    last_intensity = build_last_peak_intensity(peaks_csv, last_intensity_csv, args.chunksize)
    df = material.merge(last_intensity, on=["split", "index", "material_id"], how="left")

    # Prefer the material-level last peak position generated by the original stats.
    df["last_peak_intensity"] = pd.to_numeric(df["last_peak_intensity"], errors="coerce")
    df[["split", "index", "material_id", "last_peak_2theta", "last_peak_intensity", "min_peak_gap"]].to_csv(
        args.output_dir / "mp20_xrd_distribution_materials.csv", index=False
    )

    summaries = [
        describe(df["last_peak_2theta"], "last_peak_2theta"),
        describe(df["last_peak_intensity"], "last_peak_intensity"),
        describe(df["min_peak_gap"], "per_material_min_peak_gap"),
    ]
    pd.DataFrame(summaries).to_csv(args.output_dir / "distribution_summary.csv", index=False)

    save_hist(
        df["last_peak_2theta"],
        args.output_dir / "last_peak_2theta_distribution.png",
        title="Distribution of Last XRD Peak Position",
        xlabel="Last peak 2theta (degree)",
        bins=np.linspace(120, 180, 121),
        color="#2563eb",
    )
    save_hist(
        df["last_peak_intensity"],
        args.output_dir / "last_peak_intensity_distribution.png",
        title="Distribution of Last XRD Peak Relative Intensity",
        xlabel="Last peak relative intensity (max peak = 100)",
        bins=np.linspace(0, 100, 101),
        color="#059669",
    )
    save_joint(df, args.output_dir / "last_peak_2theta_vs_intensity.png")
    positive_gap = df.loc[df["min_peak_gap"] > 0, "min_peak_gap"]
    gap_min = max(float(positive_gap.min()), 1e-6)
    gap_max = float(positive_gap.quantile(0.999))
    save_hist(
        positive_gap,
        args.output_dir / "min_peak_gap_distribution.png",
        title="Distribution of Per-Material Minimum Adjacent Peak Gap",
        xlabel="Minimum adjacent peak gap per material (degree 2theta, log scale)",
        bins=np.logspace(np.log10(gap_min), np.log10(gap_max), 90),
        log_x=True,
        xlim=(gap_min, gap_max),
        color="#7c3aed",
    )
    write_readme(args.output_dir / "README.md")

    print(f"Wrote figures and summary to {args.output_dir}")


if __name__ == "__main__":
    main()