DM-34264: Design Task for applying real-bogus scoring #4
Conversation
| """ Constructor | ||
| """ | ||
| super(dummyClassifier, self).__init__() | ||
| super().__init__() |
There was a problem hiding this comment.
Why remove the docstring? You wanted to suggest corrections?
There was a problem hiding this comment.
No docstrings on __init__ methods, and docstrings should contain information you wouldn't otherwise get from the method name.
https://developer.lsst.io/python/numpydoc.html#placement-of-class-docstrings
| @@ -21,9 +19,6 @@ def __init__(self): | |||
| self.fc2 = nn.Linear(128, 1) | |||
|
|
|||
| def forward(self, x): | |||
| """ Forward pass | |||
There was a problem hiding this comment.
Why remove the docstring? You wanted to suggest corrections?
There was a problem hiding this comment.
This docstring doesn't contain any real information. The method docstring in the parent class is more informative, so we should let the system default to that.
https://pytorch.org/docs/stable/generated/torch.nn.Module.html
The interface will need to be fleshed out and have an actual test suite to go with it (the current test will fail); this provides an API for the Task to code against.
parejkoj
force-pushed
the
tickets/DM-34264
branch
from
1814c31
to
80bf359
Compare
|
I've responed to your comments and rebased everything. Please take a look at the rest of the task as well. |
Looks good. I'll start testing the functionality after this is merged, and capture potential needed changes along the way. |
DM-34264: Design Task for applying real-bogus scoring
This implements a working Task (tested with a mock of
interface.inter), and what should be a viable API for the model interface.The weights loading is currently a hack, using a config field. We won't want that in production, but until we have a better sense of the size of the weights file (both on disk and in-memory), we won't know whether we will need this to be in a separate process (e.g. a REST API server somewhere else), or how we will keep the pytorch model instantiated in memory between jobs. This at least should be enough to let us start testing real/bogus in a working pipeline.