How to save train, test data and model preditctions?

[johnny12150]

Is there a quick way to save the train, test data after splitting them?
How to preserve the model predictions and ground truth at the testing phase?

[chenyushuo]

For the first question, we suggest that you can use pickle to dump these split data, just like this:

import pickle
with open('split_data.pth', 'wb') as f:
    pickle.dump((train_data, test_data), f)

And we added save funtion for split data in #760 recently. You can use save_split_dataloaders and load_split_dataloaders to save and load split data in the later version.

For the second question, you can see these code (mainly based on #506):

import numpy as np
import torch

from recbole.data.dataloader.general_dataloader import GeneralFullDataLoader
from recbole.data.dataloader.sequential_dataloader import SequentialFullDataLoader


uid_series = np.array([1, 2])

# We assume you have load test_data and model
uid_field = test_data.dataset.uid_field
dataset = test_data.dataset
model.eval()

if isinstance(test_data, GeneralFullDataLoader):
    index = np.isin(test_data.user_df[uid_field].numpy(), uid_series)
    input_interaction = test_data.user_df[index]
elif isinstance(test_data, SequentialFullDataLoader):
    index = np.isin(test_data.uid_list, uid_series)
    input_interaction = test_data.augmentation(
        test_data.item_list_index[index], test_data.target_index[index], test_data.item_list_length[index]
    )
else:
    raise NotImplementedError

# Get scores of all items
try:
    scores = model.full_sort_predict(input_interaction)
except NotImplementedError:
    input_interaction = input_interaction.repeat(dataset.item_num)
    input_interaction.update(test_data.get_item_feature().repeat(len(uid_series)))
    scores = model.predict(input_interaction)

scores = scores.view(-1, dataset.item_num)  # scores of all item, shape: len(uid_series) * 

# Get Groud truth interaction
index = np.isin(test_data.dataset.inter_feat[uid_field].numpy(), uid_series)
real_inter = test_data.dataset.inter_feat[index]  # the ground truth interaction

[chenyushuo]

Yes, run_recbole() is just a simple function to run models. In fact, you can modify this function as your need.

[johnny12150]

@chenyushuo
I am facing an error that test_data.augmentation in the script is missing one argument.
Should I just pass uid_series in the first position?

[johnny12150]

Also, is it possible no to do data augmentation in the testing phase to generate input_interaction?

[johnny12150]

Besides, the interaction should be a dictionary of tensors rather than NumPy arrays?

Hence, I am getting the error of this.

Traceback (most recent call last):
  File "D:/codes/paper codes/Recbole/run_custom.py", line 39, in get_pred
    scores = model.full_sort_predict(input_interaction)
  File "C:\Users\wade\anaconda3\envs\pytorch\lib\site-packages\recbole\model\sequential_recommender\gru4rec.py", line 117, in full_sort_predict
    seq_output = self.forward(item_seq, item_seq_len)
  File "C:\Users\wade\anaconda3\envs\pytorch\lib\site-packages\recbole\model\sequential_recommender\gru4rec.py", line 78, in forward
    item_seq_emb = self.item_embedding(item_seq)
  File "C:\Users\wade\anaconda3\envs\pytorch\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "C:\Users\wade\anaconda3\envs\pytorch\lib\site-packages\torch\nn\modules\sparse.py", line 124, in forward
    return F.embedding(
  File "C:\Users\wade\anaconda3\envs\pytorch\lib\site-packages\torch\nn\functional.py", line 1852, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
TypeError: embedding(): argument 'indices' (position 2) must be Tensor, not numpy.ndarray

[chenyushuo]

The argument of test_data.augmentation has been updated in #559. We suggest you to use the lastest version of RecBole.
Beside, in RecBole, data augmentation is necessary in the testing phase to generate input_interaction for sequential model.
As for the interaction, it should be a dictionary of torch.tensors, you can see this page for more infomation about Interaction.