Skip to content

Archive of the Future Frame's open source Python package with inference code for foundation tabular models.

License

Notifications You must be signed in to change notification settings

edadaltocg/futureframe

Repository files navigation

💠 Future Frame

Empowering Data Scientists with Foundation Models for Tabular Data

  • This Python package allows you to interact with pre-trained foundation models for tabular data.
  • Easily fine-tune them on your classification and regression use cases in a single line of code.

Installation

pip install futureframe

Quick Start

Use Future Frame to fine-tune a pre-trained foundation model on a classification task.

# Import standard libraries
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score

import futureframe as ff

# Import data
dataset_name = "https://raw.githubusercontent.com/futureframeai/futureframe/main/tests/data/churn.csv"
target_variable = "Churn"
df = pd.read_csv(dataset_name)

X, y = df.drop(columns=[target_variable]), df[target_variable]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)

# Fine-tune a pre-trained classifier with Future Frame
model = ff.models.cm2.CM2Classifier()
model.finetune(X_train, y_train)

y_pred = model.predict(X_test)
auc = roc_auc_score(y_test, y_pred)
print(f"AUC: {auc:0.2f}")

Models

Model Name Paper Title Paper GitHub
CM2 Towards Cross-Table Masked Pretraining for Web Data Mining Ye et al., 2024 Link
CARTE (soon!) CARTE: Pretraining and Transfer for Tabular Learning Kim et al., 2024 Link
TabText (soon!) TabText: A Flexible and Contextual Approach to Tabular Data Representation Carballo et al., 2023 -
TabPFN (soon!) TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second Hollmann et al., 2022 Link
TransTab (soon!) Transtab: Learning Transferable Tabular Transformers Across Tables Wang et al., 2022 Link

More models will be integrated into the library soon!

More to come!

Important links

Contributing

  • We are currently under heavy development.
  • To report a bug, please write an issue.

About

Archive of the Future Frame's open source Python package with inference code for foundation tabular models.

Resources

License

Stars

Watchers

Forks

Contributors 3

  •  
  •  
  •  

Languages