Stephen Bonner

Passing a dropout value to a model through the pipeline API

5

It would be great to be able to pass a dropout value to all models, thus enabling dropout, via the pipeline api, rather than having to create a new `ERModel`...

Adding support for Apple Silicon MPS Devices (M1/Pro/Max/Ultra) WIP

15

### Description of the Change Since PyTorch has added support for Apple Silicon machines (via the use of Metal Performance Shaders https://pytorch.org/blog/introducing-accelerated-pytorch-training-on-mac/), I thought it would be interesting to add...

Understanding controlling the random seed using the hpo_pipeline

6

Hi, I am trying to understand how to control the setting of the random seed using the `hpo_pipeline` object and what exactly it is controlling. As it stands, I have...

Is there anyway to resume an ablation study?

4

Hi All, I was wondering is there anyway to resume an ablation study if it gets interrupted for any reason? It seems a shame to have to repeat all the...

KGE can not work with python 3.8 + pytorch 1.7

6

@Hi, I am trying to run the example from the readme as follows: `DGLBACKEND=pytorch dglke_train --model_name TransE_l2 --dataset FB15k --batch_size 1000 \ --neg_sample_size 200 --hidden_dim 400 --gamma 19.9 --lr 0.25...

fix to pass any negative sampler kwags through if present

1

### Link to the relevant Bug(s) Closes #1118 ### Description of the Change There is now a check for `negative_sampler_kwargs` which sets them even if there is no `negative_sampler` value....

### Describe the bug It seems that any `negative_sampler_kwargs` are ignored unless a sampler is explicitly set in the pipeline. For example, setting a value for the number of negative...

Updating the readme with MacOS install tips.

2

Adding a few tips I needed to get this to run on an Intel based MacOS install. I haven't tested on my M1 machine yet but will update further if...

Add option to reweight loss to account for differences in relation frequency

7

Sharing some initial work I had done around adding support for reweighting the loss to account for relation frequency. The current code is very much work in progress but I...

The function from_networkx is very slow when edge properties are present

6

I have been using from_networkx on a relatively large graph (45K vertecies and 2M edges) without runtime issues - the conversion happens within a minute or so. However, as soon...