-
The following preprocessing steps can be quite tedious. Please post issues if you cannot run the scripts.
-
datasets: Amazon
-- Rating file inFiles/Small subsets for experimentation
-- Meta files inPer-category files
, [metadata], [image features]
There has been an issue with the dataset site lately,
as it automatically redirects to an updated version of the dataset.
Keep pressing ESC
to stop the redirecting action.
- Performing 5-core filtering, re-indexing -
run 0rating2inter.ipynb
- Train/valid/test data splitting -
run 1spliting.ipynb
- Reindexing feature IDs with generated IDs in step 1 -
run 2reindex-feat.ipynb
- Encoding text/image features -
run 3feat-encoder.ipynb
- Filling your data description file
*.yaml
undersrc/configs/dataset
with the generated file names*.inter
,*-feat.npy
, etc. - Specifying your evaluated dataset by cmd:
python -d sports -m BM3
.
- Run
dualgnn-gen-u-u-matrix.py
on a datasetbaby
:
python dualgnn-gen-u-u-matrix.py -d baby
- The generated u-u graph should be located in the same dir as the dataset.