Yihe Dong

Hi @giuliacassara, happy to help. We originally did not include these data files, as they were too large to host on Github. I found the [script we used for converting](https://gist.github.com/twistedcubic/01a754b4984cec537a6a9ff06dc4c3a5)...

Hello @wwrlearner, yes we were surprised by the results of HyperGCN when we benchmarked it as well. [This is the code](https://github.com/twistedcubic/HNHN/blob/master/baseline.py#L101) for training HyperGCN we used. We tuned its various...

Hi @aeroplanepaper, that is interesting, I just pulled the codebase again and ran it out of the box, using [`args = gen_data_cora(args, data_path=data_path, do_val=True)`](https://github.com/twistedcubic/HNHN/blob/43132e38b3fbdd74425f359b425ceb97bd795a48/hypergraph.py#L674), and consistently got ~6% standard deviation...

Hi @aeroplanepaper, that is indeed surprising. What parameter are you varying that led to this variance? And on which dataset? Based on what I tested, previously (by re-pulling this repo)...

Hi @mayyxeng, We are glad you are using our code. The partition files are produced by the [KaHIP](https://github.com/KaHIP/KaHIP) graph partitioner. Can you try installing it? It's also a good idea...

Hi @bianzheng123, Glad you are running the code and are working with the SIFT dataset. There are several factors that affect the training loss. The key metric we used is...

Hi Navid, The `_multiprobe` module refers to the python module we built that implements the cross polytope LSH baseline. The code lives in a (non-obvious) [branch of `Falconn`](https://github.com/FALCONN-LIB/FALCONN/tree/multiprobe_test/src/multiprobe_test). The module...

Hi @pzque, We are glad you found this, it looks like the ITQ code was accidentally removed. It has been added back.

Hi Thomas, Thanks for running our code. You must be encountering the [exception on this line](https://github.com/twistedcubic/learn-to-hash/blob/master/utils.py#L126). sift is indeed enabled by default [here](https://github.com/twistedcubic/learn-to-hash/blob/master/utils.py#L80). We can turn off sift with `--sift...