File size: 14,758 Bytes
d4c177b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 | #!/usr/bin/env python3
"""
explore_dataset.py
Exploratory Data Analysis (EDA) for the CS/ML Academic Papers Dataset.
Loads the locally-saved dataset (or downloads from HuggingFace Hub), computes
summary statistics, identifies top terms per category via TF-IDF, and saves
publication-ready visualisations as PNG files.
Usage
-----
# Analyse the local dataset
python explore_dataset.py
# Load from the HuggingFace Hub instead
python explore_dataset.py --from-hub gr8monk3ys/cs-ml-academic-papers
# Customise the output directory for plots
python explore_dataset.py --plots-dir ./plots
"""
from __future__ import annotations
import argparse
import logging
import textwrap
from collections import Counter
from pathlib import Path
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from datasets import DatasetDict, load_from_disk
from sklearn.feature_extraction.text import TfidfVectorizer
# Use non-interactive backend so the script works headlessly.
matplotlib.use("Agg")
LOG = logging.getLogger("explore_dataset")
DATA_DIR: Path = Path(__file__).resolve().parent / "data"
PLOTS_DIR: Path = Path(__file__).resolve().parent / "plots"
# Colour palette (colour-blind friendly, adapted from Tol's muted scheme).
PALETTE = ["#332288", "#88CCEE", "#44AA99", "#117733", "#999933",
"#DDCC77", "#CC6677", "#882255", "#AA4499"]
# ---------------------------------------------------------------------------
# Loading helpers
# ---------------------------------------------------------------------------
def load_local(data_dir: Path) -> pd.DataFrame:
"""Load the dataset from a local ``save_to_disk`` directory."""
ds_path = data_dir / "hf_dataset"
if not ds_path.exists():
raise FileNotFoundError(
f"No saved dataset found at {ds_path}. "
"Run create_dataset.py first or use --from-hub."
)
dd = load_from_disk(str(ds_path))
frames = [dd[split].to_pandas() for split in dd]
return pd.concat(frames, ignore_index=True)
def load_hub(repo_id: str) -> pd.DataFrame:
"""Download the dataset from the HuggingFace Hub."""
from datasets import load_dataset
dd = load_dataset(repo_id)
frames = [dd[split].to_pandas() for split in dd]
return pd.concat(frames, ignore_index=True)
# ---------------------------------------------------------------------------
# Statistics
# ---------------------------------------------------------------------------
def print_summary(df: pd.DataFrame) -> None:
"""Print high-level summary statistics to stdout."""
separator = "=" * 60
print(f"\n{separator}")
print(" CS/ML Academic Papers Dataset — Summary Statistics")
print(separator)
print(f"\n Total papers : {len(df):,}")
print(f" Unique arXiv IDs : {df['arxiv_id'].nunique():,}")
print(f" Unique primary cats : {df['primary_category'].nunique()}")
# Date range
if "published" in df.columns:
dates = pd.to_datetime(df["published"], errors="coerce")
valid = dates.dropna()
if len(valid) > 0:
print(f" Published date range : {valid.min():%Y-%m-%d} to {valid.max():%Y-%m-%d}")
# Authors
author_counts = df["authors"].apply(len)
print(f"\n Authors per paper (mean): {author_counts.mean():.1f}")
print(f" Authors per paper (med) : {author_counts.median():.0f}")
# Abstract lengths
abs_len = df["abstract"].str.split().str.len()
print(f"\n Abstract length (words):")
print(f" mean : {abs_len.mean():.0f}")
print(f" median : {abs_len.median():.0f}")
print(f" min : {abs_len.min():.0f}")
print(f" max : {abs_len.max():.0f}")
print(f" std : {abs_len.std():.1f}")
# Category distribution
print(f"\n Primary category distribution:")
for cat, count in df["primary_category"].value_counts().items():
pct = 100.0 * count / len(df)
print(f" {cat:<12s} {count:>5,} ({pct:5.1f}%)")
# DOI availability
has_doi = (df["doi"].str.len() > 0).sum()
print(f"\n Papers with DOI : {has_doi:,} ({100*has_doi/len(df):.1f}%)")
print(f"\n{separator}\n")
# ---------------------------------------------------------------------------
# TF-IDF keyword extraction
# ---------------------------------------------------------------------------
def top_tfidf_terms(
df: pd.DataFrame,
text_col: str = "abstract",
group_col: str = "primary_category",
top_n: int = 15,
) -> dict[str, list[tuple[str, float]]]:
"""
For each group in *group_col*, fit a TF-IDF vectoriser on the documents
belonging to that group and return the top-*n* terms by mean TF-IDF score.
"""
results: dict[str, list[tuple[str, float]]] = {}
vectorizer = TfidfVectorizer(
max_features=5000,
stop_words="english",
min_df=5,
max_df=0.85,
ngram_range=(1, 2),
token_pattern=r"(?u)\b[a-zA-Z][a-zA-Z+#\-]{2,}\b",
)
for group, sub_df in df.groupby(group_col):
texts = sub_df[text_col].tolist()
if len(texts) < 10:
LOG.warning("Skipping group %s — too few documents (%d).", group, len(texts))
continue
tfidf_matrix = vectorizer.fit_transform(texts)
mean_scores = np.asarray(tfidf_matrix.mean(axis=0)).flatten()
feature_names = vectorizer.get_feature_names_out()
top_indices = mean_scores.argsort()[::-1][:top_n]
results[group] = [
(feature_names[i], float(mean_scores[i])) for i in top_indices
]
return results
def print_top_terms(terms_by_cat: dict[str, list[tuple[str, float]]]) -> None:
"""Pretty-print TF-IDF top terms per category."""
print("=" * 60)
print(" Top TF-IDF Terms per Category")
print("=" * 60)
for cat in sorted(terms_by_cat):
print(f"\n [{cat}]")
for rank, (term, score) in enumerate(terms_by_cat[cat], 1):
print(f" {rank:>2}. {term:<30s} (score: {score:.4f})")
print()
# ---------------------------------------------------------------------------
# Visualisations
# ---------------------------------------------------------------------------
def _savefig(fig: plt.Figure, path: Path) -> None:
fig.savefig(str(path), dpi=150, bbox_inches="tight", facecolor="white")
plt.close(fig)
LOG.info("Saved plot -> %s", path)
def plot_category_distribution(df: pd.DataFrame, output_dir: Path) -> None:
"""Bar chart of primary-category counts."""
counts = df["primary_category"].value_counts().sort_values(ascending=True)
fig, ax = plt.subplots(figsize=(8, 5))
bars = ax.barh(counts.index, counts.values, color=PALETTE[: len(counts)])
ax.bar_label(bars, padding=4, fontsize=9)
ax.set_xlabel("Number of Papers")
ax.set_title("Papers by Primary arXiv Category")
ax.spines[["top", "right"]].set_visible(False)
fig.tight_layout()
_savefig(fig, output_dir / "category_distribution.png")
def plot_abstract_length_histogram(df: pd.DataFrame, output_dir: Path) -> None:
"""Histogram of abstract word counts."""
lengths = df["abstract"].str.split().str.len()
fig, ax = plt.subplots(figsize=(8, 5))
ax.hist(lengths, bins=50, color=PALETTE[0], edgecolor="white", alpha=0.85)
ax.axvline(lengths.median(), color=PALETTE[6], linestyle="--", linewidth=1.5,
label=f"Median ({lengths.median():.0f} words)")
ax.axvline(lengths.mean(), color=PALETTE[4], linestyle=":", linewidth=1.5,
label=f"Mean ({lengths.mean():.0f} words)")
ax.set_xlabel("Abstract Length (words)")
ax.set_ylabel("Frequency")
ax.set_title("Distribution of Abstract Lengths")
ax.legend(frameon=False)
ax.spines[["top", "right"]].set_visible(False)
fig.tight_layout()
_savefig(fig, output_dir / "abstract_length_histogram.png")
def plot_abstract_length_by_category(df: pd.DataFrame, output_dir: Path) -> None:
"""Box plot of abstract lengths grouped by primary category."""
df = df.copy()
df["abstract_words"] = df["abstract"].str.split().str.len()
cats = df["primary_category"].value_counts().index.tolist()
data = [df.loc[df["primary_category"] == c, "abstract_words"].values for c in cats]
fig, ax = plt.subplots(figsize=(8, 5))
bp = ax.boxplot(data, labels=cats, patch_artist=True, showfliers=False)
for patch, colour in zip(bp["boxes"], PALETTE):
patch.set_facecolor(colour)
patch.set_alpha(0.7)
ax.set_ylabel("Abstract Length (words)")
ax.set_title("Abstract Length by Primary Category")
ax.spines[["top", "right"]].set_visible(False)
fig.tight_layout()
_savefig(fig, output_dir / "abstract_length_by_category.png")
def plot_authors_per_paper(df: pd.DataFrame, output_dir: Path) -> None:
"""Histogram of author counts per paper."""
author_counts = df["authors"].apply(len)
fig, ax = plt.subplots(figsize=(8, 5))
max_display = int(author_counts.quantile(0.99)) + 1
ax.hist(
author_counts.clip(upper=max_display),
bins=range(1, max_display + 2),
color=PALETTE[2],
edgecolor="white",
alpha=0.85,
align="left",
)
ax.set_xlabel("Number of Authors")
ax.set_ylabel("Frequency")
ax.set_title("Authors per Paper")
ax.spines[["top", "right"]].set_visible(False)
fig.tight_layout()
_savefig(fig, output_dir / "authors_per_paper.png")
def plot_publication_timeline(df: pd.DataFrame, output_dir: Path) -> None:
"""Monthly publication counts over time."""
dates = pd.to_datetime(df["published"], errors="coerce").dropna()
monthly = dates.dt.to_period("M").value_counts().sort_index()
fig, ax = plt.subplots(figsize=(10, 5))
ax.bar(
range(len(monthly)),
monthly.values,
color=PALETTE[1],
edgecolor="white",
width=1.0,
)
# Show a subset of tick labels to avoid crowding.
step = max(1, len(monthly) // 12)
tick_indices = list(range(0, len(monthly), step))
ax.set_xticks(tick_indices)
ax.set_xticklabels(
[str(monthly.index[i]) for i in tick_indices],
rotation=45,
ha="right",
fontsize=8,
)
ax.set_xlabel("Month")
ax.set_ylabel("Number of Papers")
ax.set_title("Publication Timeline (Monthly)")
ax.spines[["top", "right"]].set_visible(False)
fig.tight_layout()
_savefig(fig, output_dir / "publication_timeline.png")
def plot_top_terms_heatmap(
terms_by_cat: dict[str, list[tuple[str, float]]],
output_dir: Path,
top_n: int = 10,
) -> None:
"""Heatmap-style visualisation of top TF-IDF terms across categories."""
# Gather the union of top terms across all categories.
all_terms: list[str] = []
for cat in sorted(terms_by_cat):
for term, _ in terms_by_cat[cat][:top_n]:
if term not in all_terms:
all_terms.append(term)
cats = sorted(terms_by_cat.keys())
matrix = np.zeros((len(all_terms), len(cats)))
for j, cat in enumerate(cats):
term_map = dict(terms_by_cat[cat])
for i, term in enumerate(all_terms):
matrix[i, j] = term_map.get(term, 0.0)
fig, ax = plt.subplots(figsize=(10, max(6, 0.35 * len(all_terms))))
im = ax.imshow(matrix, aspect="auto", cmap="YlOrRd", interpolation="nearest")
ax.set_xticks(range(len(cats)))
ax.set_xticklabels(cats, fontsize=9)
ax.set_yticks(range(len(all_terms)))
ax.set_yticklabels(all_terms, fontsize=8)
ax.set_title("Top TF-IDF Terms by Category")
fig.colorbar(im, ax=ax, label="Mean TF-IDF Score", shrink=0.6)
fig.tight_layout()
_savefig(fig, output_dir / "tfidf_terms_heatmap.png")
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(
description="Exploratory Data Analysis for the CS/ML Academic Papers Dataset.",
)
parser.add_argument(
"--data-dir",
type=str,
default=str(DATA_DIR),
help=f"Local data directory (default: {DATA_DIR}).",
)
parser.add_argument(
"--from-hub",
type=str,
default=None,
help="Load the dataset from a HuggingFace Hub repo instead of locally.",
)
parser.add_argument(
"--plots-dir",
type=str,
default=str(PLOTS_DIR),
help=f"Directory for saved plots (default: {PLOTS_DIR}).",
)
parser.add_argument(
"--verbose",
action="store_true",
help="Enable debug logging.",
)
return parser.parse_args()
def main() -> None:
args = parse_args()
logging.basicConfig(
level=logging.DEBUG if args.verbose else logging.INFO,
format="%(asctime)s %(levelname)-8s %(name)s %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
)
# ------------------------------------------------------------------
# 1. Load data
# ------------------------------------------------------------------
if args.from_hub:
LOG.info("Loading dataset from HuggingFace Hub: %s", args.from_hub)
df = load_hub(args.from_hub)
else:
LOG.info("Loading dataset from local directory: %s", args.data_dir)
df = load_local(Path(args.data_dir))
LOG.info("Loaded %d papers.", len(df))
# ------------------------------------------------------------------
# 2. Summary statistics
# ------------------------------------------------------------------
print_summary(df)
# ------------------------------------------------------------------
# 3. TF-IDF keyword extraction
# ------------------------------------------------------------------
LOG.info("Computing TF-IDF top terms per category ...")
terms_by_cat = top_tfidf_terms(df)
print_top_terms(terms_by_cat)
# ------------------------------------------------------------------
# 4. Visualisations
# ------------------------------------------------------------------
plots_dir = Path(args.plots_dir)
plots_dir.mkdir(parents=True, exist_ok=True)
LOG.info("Generating visualisations -> %s", plots_dir)
plot_category_distribution(df, plots_dir)
plot_abstract_length_histogram(df, plots_dir)
plot_abstract_length_by_category(df, plots_dir)
plot_authors_per_paper(df, plots_dir)
plot_publication_timeline(df, plots_dir)
if terms_by_cat:
plot_top_terms_heatmap(terms_by_cat, plots_dir)
LOG.info("All plots saved to %s", plots_dir)
print(f"Visualisations saved to: {plots_dir}")
if __name__ == "__main__":
main()
|