running-chollet-ratio-check

What problem does it solve?

This Skill helps you choose the right model family for text and sequence classification by checking whether your dataset is better suited to a linear baseline, a small neural model, or a pretrained Transformer.

Core Features & Use Cases

• Computes the Chollet samples-per-word ratio from the training set and uses it to separate low-data, transition, and high-data regimes.
• Recommends concrete text-classification starting points such as TF-IDF plus logistic regression, small 1D-CNN or LSTM models, or pretrained Transformer fine-tuning.
• Guards against common mistakes by checking per-class sample counts, adjusting for domain-specific pretraining, and refusing to apply the heuristic to tabular, image, audio, or generative tasks.
• Requires a mandatory TF-IDF and linear baseline comparison before any deep model is treated as the final answer.
• Use case: a team deciding whether to fine-tune BERT on a small intent dataset can use this Skill to justify a linear baseline first and avoid overfitting.

Quick Start

Ask the assistant to evaluate your training text data with the Chollet ratio and return the recommended model family, baseline comparison plan, and regularization defaults.