Seeking Clarification on Model Training Process - Unable to Reproduce Satisfactory Results with Current Configuration

[SharedyX]

Thank you for sharing this work! I’ve tried replicating the training process using the documented settings (hyperparameters, dataset, etc.), but the model consistently fails to achieve satisfactory performance even after multiple attempts. Could you confirm if there are any undocumented steps (e.g., data augmentation, initialization, or code adjustments) critical for successful training? Any guidance would be greatly appreciated!

[harper-233]

Hi, thank you for raising this issue! I'm facing a similar problem while trying to replicate the paper's results on the OpTC dataset. Since the authors didn't provide the word2vec model parameters, I trained a word2vec model using the 6 days of benign data mentioned in the paper (with code origin settings). For the other model components, I used the authors' publicly available pre-trained parameters. However, my results fail to achieve satisfactory performance
Have you managed to resolve this issue or improve the performance? Could you share any insights on potential adjustments, such as data preprocessing, hyperparameter tuning, or model initialization? Any advice would be greatly appreciated!

[Wajihulhassan]

@Mati607