model.py

import os
from pathlib import Path

import numpy as np
import pandas as pd
import torch
from matplotlib import pyplot as plt
from pytorch_tabnet.pretraining import TabNetPretrainer
from pytorch_tabnet.tab_model import TabNetClassifier
from sklearn.metrics import roc_auc_score
from sklearn.preprocessing import LabelEncoder

# -- settings
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")

# TabNetPretrainer
unsupervised_model = TabNetPretrainer(
    optimizer_fn=torch.optim.Adam,
    optimizer_params=dict(lr=1e-3),
    mask_type='entmax', # "sparsemax",
    device_name='cuda'
)

clf = TabNetClassifier(
    optimizer_fn=torch.optim.Adam,
    optimizer_params=dict(lr=1e-3),
    scheduler_params={"step_size":7, # how to use learning rate scheduler
                      "gamma":0.9},
    scheduler_fn=torch.optim.lr_scheduler.StepLR,
    mask_type='entmax', # This will be overwritten if using pretrain model,
    device_name='cuda'

)