|
--- |
|
annotations_creators: [] |
|
language: [] |
|
language_creators: [] |
|
license: [] |
|
multilinguality: [] |
|
pretty_name: diabetes-readmission |
|
size_categories: |
|
- 100K<n<1M |
|
source_datasets: [] |
|
tags: |
|
- interpretability |
|
- fairness |
|
- medicine |
|
task_categories: |
|
- tabular-classification |
|
task_ids: [] |
|
--- |
|
|
|
Port of the diabetes-readmission dataset from UCI (link [here](https://archive.ics.uci.edu/ml/datasets/diabetes+130-us+hospitals+for+years+1999-2008)). See details there and use carefully. |
|
|
|
Basic preprocessing done by the [imodels team](https://github.com/csinva/imodels) in [this notebook](https://github.com/csinva/imodels-data/blob/master/notebooks_fetch_data/00_get_datasets_custom.ipynb). |
|
|
|
The target is the binary outcome `readmitted`. |
|
|
|
### Sample usage |
|
|
|
Load the data: |
|
|
|
``` |
|
from datasets import load_dataset |
|
|
|
dataset = load_dataset("imodels/diabetes-readmission") |
|
df = pd.DataFrame(dataset['train']) |
|
X = df.drop(columns=['readmitted']) |
|
y = df['readmitted'].values |
|
``` |
|
|
|
Fit a model: |
|
|
|
``` |
|
import imodels |
|
import numpy as np |
|
|
|
m = imodels.FIGSClassifier(max_rules=5) |
|
m.fit(X, y) |
|
print(m) |
|
``` |
|
|
|
|
|
Evaluate: |
|
|
|
|
|
``` |
|
df_test = pd.DataFrame(dataset['test']) |
|
X_test = df.drop(columns=['readmitted']) |
|
y_test = df['readmitted'].values |
|
print('accuracy', np.mean(m.predict(X_test) == y_test)) |
|
``` |