DEDUPEDWho it is forA data scientist choosing between collaborative filtering, sequential and context-aware models.
The problemComparing models from different codebases mixes up data splits and metrics.
The outcomeA table of metrics for several models trained with one framework and one split.
What you need first
- Python 3.7 or later and torch 1.7.0 or later
- Install with `pip install recbole` or `conda install -c aibox recbole`, or from source to get the scripts
- Optional: CUDA 9.2 or later for GPU training
Steps
- Install from source as in the README and run the default with
python run_recbole.py - Change the model with
python run_recbole.py --model=[model_name], replacing model_name with one from the model list - Adjust parameters with
python run_recbole.py --learning_rate=0.0001 --embedding_size=128 - Record each run's recall@10 and ndcg@10 and compare
Prompt for your agent
Using RUCAIBox/RecBole, run `python run_recbole.py --model=[model_name]` for three models of my choice on ml-100k and give me a table of recall@10 and ndcg@10 for each.
Caveat. Larger models and datasets need a GPU; the README links time and memory cost tables per model family.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for RUCAIBox/RecBole and "Compare recommendation models on the same data"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer