24-679: Pokémon Rarity Tabular Classifier

This model classifies Pokémon from the 151 generation dataset as Common (0) or Rare (1). It trains a tabular classifier with AutoGluon Tabular for the Carnegie Mellon University 24-679 course.

The intended use is teaching tabular data preprocessing, AutoML classification, and evaluation. It classifies rarity based on numeric attributes (HP, attack count, retreat cost) and categorical features (type, stage, ability).

Inputs and outputs

Supply a pandas DataFrame with features hp, attack_count, retreat_cost, type, stage, and has_ability. predict() returns 0 (Common) or 1 (Rare), and predict_proba() returns class probabilities.

Label Meaning
0 Common
1 Rare

Data and training

Source: pakiino/2026-24679-pokemon-151-tabular-hw1.

Partition Total
Training 336
Validation 4
Test 5
Training setting Recorded value
Framework AutoGluon Tabular 1.6.1
Target label
Problem type binary
Preset medium_quality
Selection metric Balanced accuracy
Fit budget 300 seconds
Best Model WeightedEnsemble_L2 (NeuralNetTorch)
Recorded environment Linux; Python 3.13.15; PyTorch 2.11.0+cpu

The course notebook reports 19.66 seconds of total training time. The best model achieved a validation score (balanced_accuracy) of 1.0 on the 4 validation rows.

Evaluation

Test metric Value
Test size 5 original responses
One error change 20.0 percentage points
Training-majority baseline predicts Common

This small test set of 5 examples is not a comprehensive evaluation. Broader use requires a larger evaluation dataset.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train VictorTishkevich/pokemon-automl-model