codeKgu

from transformers import Trainer, TrainingArguments

training_args = TrainingArguments(
    output_dir="./logs/model_name",
    logging_dir="./logs/runs",
    overwrite_output_dir=True,
    do_train=True,
    per_device_train_batch_size=32,
    num_train_epochs=1,
    evaluate_during_training=True,

codeKgu

from multimodal_transformers.model import AutoModelWithTabular, TabularConfig
from transformers import AutoConfig

num_labels = len(np.unique(torch_dataset, labels))
config = AutoConfig.from_pretrained('bert-base-uncased')
tabular_config = TabularConfig(
    num_labels=num_labels,
    cat_feat_dim=torch_dataset.cat_feats.shape[1],
    numerical_feat_dim=torch_dataset.numerical_feats.shape[1],
    combine_feat_method='weighted_feature_sum_on_transformer_cat_and_numerical_feats',

codeKgu

import pandas as pd
from multimodal_transformers.data import load_data
from transformers import AutoTokenizer

data_df = pd.read_csv('Womens Clothing E-Commerce Reviews.csv')
text_cols = ['Title', 'Review Text']
# The label col is expected to contain integers from 0 to N_classes - 1
label_col = 'Recommended IND' 
categorical_cols = ['Clothing ID', 'Division Name', 'Department Name', 'Class Name']
numerical_cols = ['Rating', 'Age', 'Positive Feedback Count']

codeKgu

self._base_seed = torch.empty((), dtype=torch.int64).random_(generator=loader.generator).item()

codeKgu

import scipy.sparse as sp
import numpy as np
 
def init_node_feats(self, type, device):
    if type == 'one_hot_init':
        num_nodes = self.graph.shape[0]
        identity = sp.identity(num_nodes)
        ind0, ind1, values = sp.find(identity)
        inds = np.stack((ind0, ind1), axis=0)
        self.node_feats = torch.sparse_coo_tensor(inds, values, device=device, dtype=torch.float)