Первичная версия: стилометрия реплик, LSA и протокол экспериментов

2026-10-24 01:17:36 +03:00
commit c390224500
52 changed files with 10099 additions and 0 deletions
--- a/lab1/stylometry/plotting.py
+++ b/lab1/stylometry/plotting.py
@@ -0,0 +1,167 @@
+"""Построение графиков для отчёта и презентации."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Sequence
+
+import matplotlib
+
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt  # noqa: E402
+import numpy as np  # noqa: E402
+import pandas as pd  # noqa: E402
+from sklearn.decomposition import PCA  # noqa: E402
+from sklearn.manifold import TSNE  # noqa: E402
+
+
+def _save(fig: plt.Figure, path: str | Path) -> Path:
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    fig.tight_layout()
+    fig.savefig(path, dpi=200, bbox_inches="tight")
+    plt.close(fig)
+    return path
+
+
+def plot_class_distribution(
+    frame: pd.DataFrame,
+    character_column: str = "character",
+    path: str | Path = "results/figures/class_distribution.png",
+) -> Path:
+    counts = frame[character_column].value_counts().sort_values()
+    fig, ax = plt.subplots(figsize=(8, max(6, len(counts) * 0.25)))
+    ax.barh(counts.index, counts.values)
+    ax.set_xlabel("Число реплик")
+    ax.set_ylabel("Персонаж")
+    ax.set_title("Распределение реплик по персонажам")
+    return _save(fig, path)
+
+
+def plot_scree(
+    explained_variance_ratio: Sequence[float],
+    path: str | Path = "results/figures/lsa_scree.png",
+    title: str = "Доля объяснённой дисперсии компонент LSA",
+) -> Path:
+    values = np.asarray(explained_variance_ratio, dtype=float)
+    cumulative = np.cumsum(values)
+    fig, ax = plt.subplots(figsize=(7, 4.5))
+    ax.plot(np.arange(1, len(values) + 1), cumulative, marker="o", markersize=3)
+    ax.set_xlabel("Число компонент")
+    ax.set_ylabel("Накопленная доля дисперсии")
+    ax.set_title(title)
+    ax.grid(alpha=0.3)
+    return _save(fig, path)
+
+
+def plot_confusion_matrix(
+    matrix: pd.DataFrame,
+    path: str | Path = "results/figures/confusion_matrix.png",
+    title: str = "Матрица ошибок",
+    annotate: bool | None = None,
+) -> Path:
+    values = matrix.to_numpy(dtype=float)
+    n_classes = values.shape[0]
+    if annotate is None:
+        annotate = n_classes <= 20
+    fig, ax = plt.subplots(figsize=(max(8, n_classes * 0.35), max(7, n_classes * 0.32)))
+    image = ax.imshow(values, cmap="Blues", aspect="auto")
+    ax.set_xticks(np.arange(n_classes))
+    ax.set_yticks(np.arange(n_classes))
+    ax.set_xticklabels(matrix.columns, rotation=90, fontsize=7)
+    ax.set_yticklabels(matrix.index, fontsize=7)
+    ax.set_xlabel("Предсказанный класс")
+    ax.set_ylabel("Истинный класс")
+    ax.set_title(title)
+    if annotate:
+        for row in range(n_classes):
+            for column in range(n_classes):
+                value = values[row, column]
+                if value:
+                    ax.text(
+                        column, row, f"{value:.2f}" if value < 1 else f"{int(value)}",
+                        ha="center", va="center", fontsize=6,
+                        color="white" if value > values.max() * 0.5 else "black",
+                    )
+    fig.colorbar(image, ax=ax, fraction=0.046, pad=0.04)
+    return _save(fig, path)
+
+
+def plot_pca(
+    matrix: np.ndarray,
+    labels: Sequence[str],
+    path: str | Path = "results/figures/pca.png",
+    title: str = "PCA признакового пространства",
+) -> Path:
+    values = np.asarray(matrix, dtype=float)
+    if values.shape[0] < 2:
+        raise ValueError("Для PCA нужно хотя бы два объекта.")
+    coordinates = PCA(n_components=2, random_state=42).fit_transform(values)
+    fig, ax = plt.subplots(figsize=(9, 7))
+    unique_labels = sorted(set(labels))
+    cmap = plt.get_cmap("tab20")
+    color_map = {label: cmap(index % 20) for index, label in enumerate(unique_labels)}
+    for label in unique_labels:
+        mask = np.asarray(labels) == label
+        ax.scatter(
+            coordinates[mask, 0], coordinates[mask, 1],
+            s=12, alpha=0.7, label=label, color=color_map[label],
+        )
+    ax.set_xlabel("Первая главная компонента")
+    ax.set_ylabel("Вторая главная компонента")
+    ax.set_title(title)
+    ax.legend(fontsize=6, ncol=2, loc="best", frameon=False)
+    return _save(fig, path)
+
+
+def plot_tsne(
+    matrix: np.ndarray,
+    labels: Sequence[str],
+    path: str | Path = "results/figures/tsne.png",
+    title: str = "t-SNE признакового пространства",
+    perplexity: float = 30.0,
+) -> Path:
+    values = np.asarray(matrix, dtype=float)
+    if values.shape[0] < 10:
+        raise ValueError("Для t-SNE нужно хотя бы десять объектов.")
+    perplexity = min(perplexity, max(5.0, values.shape[0] / 4.0))
+    coordinates = TSNE(
+        n_components=2,
+        perplexity=perplexity,
+        init="pca",
+        learning_rate="auto",
+        random_state=42,
+    ).fit_transform(values)
+    fig, ax = plt.subplots(figsize=(9, 7))
+    unique_labels = sorted(set(labels))
+    cmap = plt.get_cmap("tab20")
+    color_map = {label: cmap(index % 20) for index, label in enumerate(unique_labels)}
+    for label in unique_labels:
+        mask = np.asarray(labels) == label
+        ax.scatter(
+            coordinates[mask, 0], coordinates[mask, 1],
+            s=10, alpha=0.7, label=label, color=color_map[label],
+        )
+    ax.set_xlabel("t-SNE 1")
+    ax.set_ylabel("t-SNE 2")
+    ax.set_title(title)
+    ax.legend(fontsize=6, ncol=2, loc="best", frameon=False)
+    return _save(fig, path)
+
+
+def plot_feature_importance(
+    names: Sequence[str],
+    values: Sequence[float],
+    path: str | Path = "results/figures/feature_importance.png",
+    title: str = "Важность признаков",
+    top_n: int = 25,
+) -> Path:
+    frame = pd.DataFrame({"feature": list(names), "value": np.asarray(values, dtype=float)})
+    frame["abs_value"] = frame["value"].abs()
+    frame = frame.sort_values("abs_value", ascending=False).head(top_n).iloc[::-1]
+    fig, ax = plt.subplots(figsize=(9, max(5, len(frame) * 0.28)))
+    ax.barh(frame["feature"], frame["value"])
+    ax.set_xlabel("Вклад признака")
+    ax.set_ylabel("Признак")
+    ax.set_title(title)
+    return _save(fig, path)