Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft of Keras pickle RFC #1

Merged
merged 26 commits into from
Sep 21, 2020
Merged

Draft of Keras pickle RFC #1

Changes from 1 commit
Commits
Show all changes
26 commits
Select commit Hold shift + click to select a range
e6d424d
incorporate suggestions
adriangb Sep 2, 2020
2345c40
Update README.md
stsievert Sep 3, 2020
cb48f7f
add note aboute SaveModel
adriangb Sep 3, 2020
fc3df6a
Edit motivation
stsievert Sep 3, 2020
d3b3205
Technical edits
stsievert Sep 3, 2020
a882e8c
Edits
stsievert Sep 4, 2020
610d13f
Single file
stsievert Sep 4, 2020
a06e2ac
better wording
stsievert Sep 4, 2020
b98f2be
spell check
stsievert Sep 4, 2020
8ed3397
Link to blog.dask
stsievert Sep 4, 2020
cd6278a
Reorganizing
stsievert Sep 4, 2020
265aee4
Reorganizing
stsievert Sep 4, 2020
3ccd7c5
Add title and status
adriangb Sep 4, 2020
42d5a07
Edit title
stsievert Sep 6, 2020
a396a6e
Add link to Dask blog post
stsievert Sep 6, 2020
a7f963c
Update 20200902-pickle-for-keras.md
adriangb Sep 11, 2020
0d4d273
Clarify support for Pickle 5
adriangb Sep 11, 2020
d0a9295
fix typo
adriangb Sep 15, 2020
81d3b7b
__reduce__ -> __reduce_ex__ for PEP 574 support
adriangb Sep 15, 2020
a160d72
Update 20200902-pickle-for-keras.md
stsievert Sep 19, 2020
957c1ba
Delete metric tests
stsievert Sep 19, 2020
0c6ef25
Change metric file
stsievert Sep 19, 2020
de2d703
Model file
stsievert Sep 19, 2020
b393fe9
Update 20200902-pickle-for-keras.md
stsievert Sep 19, 2020
349b649
Define temp_ram_location
adriangb Sep 19, 2020
42e4a63
__reduce_ex__ -> __reduce__
adriangb Sep 21, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 5 additions & 8 deletions rfcs/20200902-pickle-for-keras.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -150,11 +150,9 @@ assert m1 == m2 # TODO: or some other check
For `tf.keras.Model`, we can use `SaveModel` as the backend for `__reduce_ex__`:

``` python
# tensorflow/python/.../training.py
from tf.keras.models import load_model

class Model:
...
class NewModel(Model):
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm curious as to why the name change? I was envisioning these code blocks as representing "here's a pseudocode of what this would look like if implemented in TF" and not necessarily "here's how users can create a picklable Model" which is the first thought that came to mind when I saw NewModel(Model).

def __reduce_ex__(self, protocol):
self.save(f"ram://tmp/saving/{id(self)")
b = tf.io.gfile.read_folder(f"ram://tmp/saving/{id(self)}")
Expand All @@ -166,15 +164,14 @@ class Model:
return load_model(temp_ram_location)
adriangb marked this conversation as resolved.
Show resolved Hide resolved
```

Small augmentations to TensorFlow's `io` module would be required, as discussed in [tensorflow#39609].

By wrapping the pickled object within a `Numpy` array, pickling will support
pickle protocol 5 for zero-copy pickling. This provides an immediate
performance improvement for many use cases.

This almost exactly mirrors the PyTorch implementation of Pickle support in [pytorch#9184]
performance improvement for many use cases. This almost exactly mirrors the PyTorch
implementation of Pickle support in [pytorch#9184]
as mentioned in "[Pickle isn't slow, it's a protocol]."

Small augmentations to TensorFlow's `io` module would be required, as discussed in [tensorflow#39609].

[pytorch#9184]:https://github.com/pytorch/pytorch/pull/9184

### Alternatives Considered
Expand Down