How am I going to deal with all of these cats
summary
Practical comparisons and concrete tools appear: CatBoost ordered target encoding uses randomized running means to avoid leakage and reduce overfitting, GBM and LightGBM offer categorical splits with parameters like cat_smooth/max_cat_to_onehot, and XGBoost requires preprocessed numeric inputs. Encoder libraries mentioned include category_encoders and dirty_cat similarity encoder, fuzzywuzzy and rexpi help cleaning, and glove vectors provide pretrained embeddings. One‑hot encoding can explode to tens of thousands of columns, permutation importance finds real impact, and external data sometimes gives an order‑of‑magnitude lift.
AI-written summary · about
chapters (7)
- 0:00Handling categorical variables
- 2:23Encoding strategies
- 18:56Cleaning categories
- 27:13Choosing encoding
- 31:08Gradient boosting and categories
- 36:23Final thoughts
- 36:47Q&A
related talks
38mCatBoost - the new generation of gradient boosting
37mGradient Boosting for Data with Both Numerical and Text Features
24mThe art of categorical encoding for Tabular data problems
01h 27mCatBoost - The New Generation of Gradient Boosting
01h 33mCrunching Your Data with CatBoost - the New Gradient Boosting Library
35mUsing Embedding Layers to Manage High Cardinality Categorical Data
18mCompetitive Analysis of the Top Gradient...
29mDirty Data Science Machine Learning On Non Curated Data