DIEN model data error.

[Jeriousman]

I tried to operate autopsy on DIEN model. To do that, I need to get interaction data from dataloader. So what I did as below

import argparse

from recbole.quick_start import run_recbole

import logging
from logging import getLogger
import torch
import pickle
import recbole
from recbole.config import Config
from recbole.data import create_dataset, data_preparation, save_split_dataloaders, load_split_dataloaders, create_samplers
from recbole.utils import init_logger, get_model, get_trainer, init_seed, set_color
recbole.__file__

config_dict = {
               'load_col':
                   {'inter': ['user_id', 'item_id',  'timestamp'], ##'rating'
                    'item': ['item_id','movie_title', 'release_year', 'class'],
                    'user': ['user_id', 'age', 'gender', 'occupation']},
                   'USER_ID_FIELD': 'user_id',
                   'ITEM_ID_FIELD': 'item_id',
                   'TIME_FIELD': 'timestamp',
                    # 'LABEL_FIELD': 'rating',
                    # 'threshold': {'rating':4},
                    'save_dataset': True,
                    'save_dataloaders': True,
                    'normalize_all': True,
                    # 'seq_len': {'item_id': 30, 'class': 30},
                    ##training
                    'epochs':5,
                    'train_batch_size': 2048,
                    'learner':'adam',
                    'learning_rate': 0.001,
                    'neg_sampling':{'uniform': 1, 'dynamic':3},
                    'eval_step': 1,
                    'stopping_step':10, ##early-stopping
                    'clip_grad_norm': None,
                    'loss_decimal_place': 5,
                    'weight_decay': 0.1,
                    'require_pow': False,
                    ##evaluation
                    'eval_args': {'group_by': 'user', 'order':'TO', 'split': {'RS': [0.8, 0.1, 0.1]}, 'mode':'uni10'},
                    'metrics': ['AUC', 'logloss'],
                    'topk': 10,
                    'valid_metric': 'AUC',
                    'eval_batch_size': 4096,
                    'metric_decimal_place': 5, 
                    ##model specific
                    'dropout_prob': 0.2
                   }
                   
               
config = Config(model='DIEN', dataset = 'ml-100k', config_dict= config_dict) ##, config_file_list=config_file_list, config_dict=config_dict)

init_seed(config['seed'], config['reproducibility'])
# logger initialization
init_logger(config)
logger = getLogger()
logger.info(config)

# dataset filtering
dataset = create_dataset(config)

# dataset splitting
train_data, valid_data, test_data = data_preparation(config, dataset)

The result of data_preparation() is as below

train_data
Out[22]: <recbole.data.dataloader.general_dataloader.TrainDataLoader at 0x7f5de5706c10>
valid_data
Out[23]: <recbole.data.dataloader.general_dataloader.NegSampleEvalDataLoader at 0x7f5de49516d0>
test_data
Out[24]: <recbole.data.dataloader.general_dataloader.NegSampleEvalDataLoader at 0x7f5de5706bd0>

It seems to be fine, but when I do this, it says None type for training dataloader.

for i, data in enumerate(train_data):
    v = data
    break

The error is below.

Traceback (most recent call last):

  File "/tmp/ipykernel_9318/823407621.py", line 1, in <module>
    for i, data in enumerate(train_data):

  File "/home/hojun/Python/Models/Recommender_systems/RecBole/recbole/data/dataloader/abstract_dataloader.py", line 79, in __next__
    return self._next_batch_data()

  File "/home/hojun/Python/Models/Recommender_systems/RecBole/recbole/data/dataloader/general_dataloader.py", line 63, in _next_batch_data
    cur_data = self._neg_sampling(self.dataset[self.pr:self.pr + self.step])

  File "/home/hojun/Python/Models/Recommender_systems/RecBole/recbole/data/dataloader/abstract_dataloader.py", line 163, in _neg_sampling
    self.model.eval()

AttributeError: 'NoneType' object has no attribute 'eval'

1. I do not know why I get NoneType and how to fix this. Can you clarify the issue?

When I do for valid and test data,

for i, data in enumerate(test_data):
    v = data
    break

I dont get the error andI get 4 outputs. I guess the 4 outputs come from NegSampleEvalDataLoader?

Thank you in advance.

[Wicknight]

@Jeriousman
Hello, thanks for your attention to RecBole!
The NoneType here refers to the model in train_data. This problem occurs because you have not passed the model of 'DIEN' into train_data before enumerating the data. You can check our function 'trainer.fit()' and you'll find that we'll first pass the model into train_data before enumerating the data if you use dynamic negative sampling.
So you can get the model first through this function：
model = get_model(config['model'])(config, train_data.dataset).to(config['device'])
and then pass the model into train_data through this function：
train_data.get_model(model)

[Jeriousman]

Thank you for the advice!