XBNet - Xtremely Boosted Network

Boosted neural network for tabular data

XBNET that is built on PyTorch combines tree-based models with neural networks to create a robust architecture that is trained by using a novel optimization technique, Boosted Gradient Descent for Tabular Data which increases its interpretability and performance. Boosted Gradient Descent is initialized with the feature importance of a gradient boosted tree, and it updates the weights of each layer in the neural network in two steps:

• Update weights by gradient descent.
• Update weights by using feature importance of a gradient boosted tree in every intermediate layer.

Features

• Better performance, training stability and interpretability for tabular data.
• Easy to implement with rapid prototyping capabilities
• Minimum Code requirements for creating any neural network with or without boosting

Comparison with XGBOOST

XBNET VS XGBOOST testing accuracy on different datasets with no hyperparameter tuning

Installation :

pip install --upgrade git+https://github.com/tusharsarkar3/XBNet.git

Example for using

import torch
import numpy as np
import pandas as pd
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split
from XBNet.training_utils import training,predict
from XBNet.models import XBNETClassifier
from XBNet.run import run_XBNET

data = pd.read_csv('test\Iris (1).csv')
print(data.shape)
x_data = data[data.columns[:-1]]
print(x_data.shape)
y_data = data[data.columns[-1]]
le = LabelEncoder()
y_data = np.array(le.fit_transform(y_data))
print(le.classes_)

X_train,X_test,y_train,y_test = train_test_split(x_data.to_numpy(),y_data,test_size = 0.3,random_state = 0)
model = XBNETClassifier(X_train,y_train,2)

criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

m,acc, lo, val_ac, val_lo = run_XBNET(X_train,X_test,y_train,y_test,model,criterion,optimizer,32,300)
print(predict(m,x_data.to_numpy()[0,:]))

Output images

Reference

If you make use of this software for your work, we would appreciate it if you would cite us:

@misc{sarkar2021xbnet,
      title={XBNet : An Extremely Boosted Neural Network}, 
      author={Tushar Sarkar},
      year={2021},
      eprint={2106.05239},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Features to be added :

• Metrics for different requirements
• Addition of some other types of layers