Add KerasHub classification guide #1948
Conversation
gowthamkpr
requested review from
fchollet,
MarkDaoust and
pcoet
as code owners
|
Thanks for the PR @gowthamkpr!! |
|
We'll need the Python and md files for the review! |
|
I've addressed all comments. PTAL!!! Thank you!! |
| modern approaches still are built of several complex components. | ||
| Luckily, Keras provides APIs to construct commonly used components. | ||
|
|
||
| This guide demonstrates Keras' modular approach to solving image |
There was a problem hiding this comment.
NIT: Keras'->kerasHub's
|
@gowthamkpr, please make sure to run |
I already used autogen.py to generate the files. The issue is that autogen.py generates images but uses absolute paths for images. The only change I made to the .md file is to replace absolute image paths from my local file system to relative paths. |
| target_lr=self.target_lr, | ||
| hold=self.hold, | ||
| ) | ||
|
|
| hold=self.hold, | ||
| ) | ||
|
|
||
| return ops.where(step > self.total_steps, 0.0, lr) |
There was a problem hiding this comment.
Why set the LR to 0? Should it not be a nonzero min_lr value?
There was a problem hiding this comment.
no, with cosine decay the lr decays to 0 at the end of training. Actually, we wouldn't need this conditional anyway, since training can't get past total_steps. I assume this was mainly for plotting purposes (it's been copied from the old version of this guide).
Adds the KerasHub image classification guide.
The guide is copied from https://keras.io/guides/keras_cv/classification_with_keras_cv/ and adapted for KerasHub.