Tianqin-Meng/fasttext-climate-classifier

FastText Climate Classifier

FastText supervised classifier trained on newspaper data for climate content detection.

Model Performance

• Precision: 82.7%
• Recall: 84.7%
• Accuracy: 93.2%
• Climate F1: 83.7%

Confusion Matrix

	Predicted Climate	Predicted Other
Actual Climate	TP: 3846	FN: 695
Actual Other	FP: 807	TN: 17433

Files

• fasttext_climate.bin - Trained FastText model
• chunk_labels.jsonl - Training data (220K labeled chunks)
• fasttext_train.txt - Training file
• fasttext_valid.txt - Validation file
• keywords.txt - Climate/nature keywords
• evaluation_metrics.json - Full metrics
• training_config.json - Training configuration

Usage

import fasttext
from fasttext.FastText import _FastText as FastTextModel

# NumPy 2.x compatibility patch
def patched_predict(self, text, k=1, threshold=0.0, on_unicode_error='strict'):
    import warnings
    with warnings.catch_warnings():
        warnings.simplefilter("ignore")
        result = self.f.predict(text, k, threshold, on_unicode_error)
        if result:
            probs = [float(p) for p, _ in result]
            labels = [l for _, l in result]
            return tuple(labels), probs
        else:
            return (), []

FastTextModel.predict = patched_predict

# Load model
model = fasttext.load_model('fasttext_climate.bin')

# Predict
labels, probs = model.predict('carbon emissions and global warming')
print(f'Label: {labels[0]}, Probability: {probs[0]:.4f}')