[WIP] 58 add goggle #162
[WIP] 58 add goggle #162
Conversation
| **kwargs: Any, | ||
| ) -> None: | ||
| kwargs.setdefault("aggr", aggr) | ||
| super().__init__(node_dim=0, **kwargs) |
There was a problem hiding this comment.
Is this node_dim a parameter or needs to be hardcoded to 3?
There was a problem hiding this comment.
I have exposed it as a parameter now.
| **kwargs, | ||
| ) | ||
|
|
||
| def enforce_constraints(self, X_synth: torch.Tensor) -> np.ndarray: |
There was a problem hiding this comment.
Nope, will delete in next push.
tests/plugins/generic/test_goggle.py
Outdated
| assert np.mean(results) > 0.7 | ||
|
|
||
|
|
||
| # @pytest.mark.skipif(sys.platform != "linux", reason="Linux only for faster results") |
There was a problem hiding this comment.
If there are no plans for conditional generation, you should delete these comments.
There was a problem hiding this comment.
I am hoping to add conditional generation as a later date, but will delete in the next push as I haven't seen the research code for it yet.
.github/workflows/test_full.yml
Outdated
| @@ -28,6 +28,7 @@ jobs: | |||
| - name: Install dependencies | |||
| run: | | |||
| pip install -r prereq.txt | |||
| pip install .[goggle] | |||
There was a problem hiding this comment.
Why the separate install profile? Isn't google included in the default library?
There was a problem hiding this comment.
Talking with Evgeny, this seems like a better way to handle the dependencies. This issue is that torch_scatter and torch_sparse need to be installed from source with an --find-links flag or similar. These dependencies are very tricky
src/synthcity/plugins/core/plugin.py
Outdated
| @@ -548,8 +548,22 @@ def __init__(self, plugins: list, expected_type: Type, categories: list) -> None | |||
| self._available_plugins = {} | |||
| for plugin in plugins: | |||
| stem = Path(plugin).stem.split("plugin_")[-1] | |||
| self._available_plugins[stem] = plugin | |||
|
|
|||
| if stem == "goggle": | |||
There was a problem hiding this comment.
Please don't include specific plugin conditions in plugin.py, this should stay generic.
- add `all` install target - drop the goggle mark for tests and the checks from the plugin. torch-sparse requires compiling. `pip` is not aware of local computer architectures, so you need to compile when installing with pip. `conda install` can handle the computer arch, but you need to be sure you match the torch version, otherwise it will crash. For now, pip compiles the library everytime. However, the github actions are caching the `pip` stage, so it won't take forever everytime.
Description
Added synthetic data model "goggle", this version does not support conditional generation. Closes #58 .
Goggle paper: "GOGGLE: Generative Modelling for Tabular Data by Learning Relational Structure" Authors: Tennison Liu, Zhaozhi Qian, Jeroen Berrevoets, Mihaela van der Schaar
Affected Dependencies
How has this been tested?
- test_plugin_sanity
- test_plugin_name
- test_plugin_type
- test_plugin_hyperparams
- test_plugin_fit
- test_plugin_generate
- test_plugin_generate_constraints_goggle
- test_sample_hyperparams
- test_eval_performance_goggle
- test_plugin_encoding
pytest -vvvs tests/plugins/generic/test_goggle.pyChecklist