Fix type error in DropPointsByClass #113
Conversation
tests/myria3d/models/test_model.py
Outdated
| def test_model_get_batch_tensor_by_enumeration(): | ||
| with hydra.initialize(config_path="./../../../trained_model_assets/", job_name="config"): | ||
| config = hydra.compose( | ||
| config_name="proto151_V2.0_epoch_100_Myria3DV3.1.0_predict_config_V3.7.0", |
There was a problem hiding this comment.
This config_name is hardcoded. Can it be defined imported (I think it is defined in run.py or train.py).
There was a problem hiding this comment.
Thanks, looks a good idea, but I'm not sure it is easy to get DEFAULT_CONFIG_FILE from run.py as it is not in the myria3d python module (hence I suppose it is not intended to be imported)
There was a problem hiding this comment.
Inded, but consequently I do not think it is supposed to be used for testing.
In test I usually used function make_default_hydra_cfg from conftest.py, to always test for the default configuration (not the one used for inference).
There was a problem hiding this comment.
I finally managed to find how to use it (the issue was that datamodule was using hydra:runtime.cwd, so I had to override the parameter that required it
| data = torch_geometric.data.Data(x=None, y=y, idx_in_original_cloud=idx) | ||
| out_data = tt(data) | ||
| assert np.array_equal(out_data.y, np.array([0, 0, 0, 0, 1, 1])) | ||
| assert np.array_equal(out_data.idx_in_original_cloud, idx) |
There was a problem hiding this comment.
Are you testing that the idx in original cloud are not modified?
There was a problem hiding this comment.
yes, after all I'm not sure it is useful, but I did it to be consistent with the test I added for DropPointsByClass
leavauchier
force-pushed
the
fix-drop-points
branch
from
6ed6e7f to
2dac87d
Compare
|
LGTM ! |
A
TypeError: object of type 'numpy.int64' has no len()was raised in some specific use case when using DropPointsByClass with batches containing a single point (the numpy array was transformed to an integer)