[ASK] Sequential Recommender (SLi_Rec) Questions
#1.040 geöffnet am 22. Jan. 2020
Repository-Metriken
- Stars
- (17.706 Stars)
- PR-Merge-Metriken
- (Durchschn. Merge 6T 16h) (10 gemergte PRs in 30 T)
Beschreibung
Description
When we create a dataset for the predict function in SLi_Rec do we need to start with a label i.e. 1 or 0 and what timestamp should we use? If we use a future timestamp does the model provide a probability based on the timestamp? How about if we use the timestamp for NOW() or a timestamp in the past?
What are the tradeoffs/benefits for min_sequence_len and max_sequence_len in terms of network complexity and compute memory and time?
Is there a way to influence the alpha parameter that balances short-term and long-term?
What were the RAM requirements for the full Amazon dataset and the training data file size?
Any tips on providing the model with a more complex representation for Category. I see that under the hood there is an embedding happening, wondering if there is a way to include a combination of columns.