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.
pip install futureframe
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}")
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!
- We are currently under heavy development.
- To report a bug, please write an issue.