Train FastText

Overview

This skill provides guidance for training FastText text classification models, particularly when facing dual constraints like model size limits and accuracy requirements. It covers systematic experimentation strategies, hyperparameter tuning approaches, and common pitfalls to avoid.

Constraint Prioritization Strategy

When facing competing constraints (e.g., model size < X MB AND accuracy >= Y%), establish a clear strategy:

1. Identify which constraint is harder to satisfy - Accuracy is typically harder to recover after compression
2. First achieve the accuracy target with an unconstrained model
3. Then apply size reduction techniques (quantization, dimension reduction, pruning)
4. Track the accuracy-size trade-off at each compression step

Systematic Experimentation Approach

Phase 1: Quick Exploratory Runs