Sparse Text Classification with Vocabulary Limits

Shows how to pair TfidfVectorizer with TlmMilpClassifier by constraining vocabulary dimension to prevent exponential MILP solver time, visualizing the resulting ternary text weights.

import matplotlib.pyplot as plt
import numpy as np
from sklearn.datasets import fetch_20newsgroups
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.pipeline import Pipeline
from tlmnet import TlmMilpClassifier

# Fetch binary text dataset subset
categories = ["alt.atheism", "sci.space"]
newsgroups = fetch_20newsgroups(
    subset="train", categories=categories, remove=("headers", "footers", "quotes")
)

# Constrain vocabulary to top 25 terms to keep MILP dimension tractable
pipeline = Pipeline(
    [
        (
            "tfidf",
            TfidfVectorizer(max_features=25, stop_words="english", sublinear_tf=True),
        ),
        ("tlm", TlmMilpClassifier(max_features=8, C=1.0)),
    ]
)

pipeline.fit(newsgroups.data, newsgroups.target)

# Retrieve selected words and their discrete ternary weights
feature_names = pipeline.named_steps["tfidf"].get_feature_names_out()
weights = pipeline.named_steps["tlm"].coef_

# Filter down to only the non-zero features selected by the MILP solver
mask = weights != 0
active_words = feature_names[mask]
active_weights = weights[mask]

# Sort by weight for cleaner plotting
sort_idx = np.argsort(active_weights)
active_words = active_words[sort_idx]
active_weights = active_weights[sort_idx]

# Plot the ternary weights
plt.figure(figsize=(8, 4))
colors = ["#d62728" if w < 0 else "#2ca02c" for w in active_weights]
plt.barh(active_words, active_weights, color=colors, edgecolor="black")

plt.xlabel("Ternary Weight (-1 or +1)")
plt.title("Exact MILP Selected Terms (Text Classification)")
plt.xticks([-1, 0, 1])
plt.grid(axis="x", linestyle="--", alpha=0.7)
plt.tight_layout()
plt.show()