TabICL

2.1.1 verified Fri May 01 auth: no python

TabICL (Tabular In-Context Learning) is a state-of-the-art tabular foundation model for few-shot and zero-shot learning on tabular data. Current version 2.1.1, actively maintained.

pip install tabicl

Common errors

error ImportError: cannot import name 'TabICL' from 'tabicl' ↓

cause Incorrect import path; possibly using an older version or wrong module name.

fix

Use 'from tabicl import TabICL'. Ensure tabicl version 2.1.1 is installed: pip install tabicl==2.1.1

error RuntimeError: Expected input to be a pandas DataFrame ↓

cause Passing a numpy array or list instead of DataFrame.

fix

Wrap your data: data = pd.DataFrame(data). TabICL's API explicitly requires DataFrame.

Warnings

breaking TabICL v2.x requires Python >=3.10 and PyTorch >=2.0. Older versions (1.x) are incompatible and use different APIs. ↓

fix Ensure Python 3.10+ and upgrade PyTorch to 2.0+. Do not install older tabicl versions.

gotcha TabICL expects data in pandas DataFrame format with no missing values. NaN or None will cause silent errors or poor performance. ↓

fix Impute missing values before passing data. Use pandas .fillna() or similar.

Imports

TabICL
wrong
```
import tabicl
```
correct
```
from tabicl import TabICL
```
Direct import of the model class is required, not just the module.

Quickstart

Basic usage of TabICL for tabular data prediction.

import pandas as pd
from tabicl import TabICL

# Load or create a pandas DataFrame with your data
# data = pd.read_csv('your_data.csv')

# Initialize the model
model = TabICL(pretrained=True)

# Example: predict on a dataset (adjust as per your task)
# predictions = model.predict(data)
print('TabICL model loaded successfully')