Feat: Upgrade `tensorflow` to `2.16.1` and `keras` to `3.4.0` #1385

RollerKnobster · 2024-06-26T08:45:34Z

To accommodate the upgrade, next changes to the code were made:

tensorflow no longer has keras wrappers, which are now used from scikeras library, recommended for use by them. They have largely the same interface, but slightly different innerworkings, which were accomodated for.
Since keras 3.* now clones the input regressor that is provided to the BaseEstimator into a regressor_ attribute, and ONLY updates the regressor_ one after fitting, metadata extraction in gordo.builder.build_model.ModelBuilder was changed to seek the regressor_ in the steps.
Accessing keras.optimizers has changed in keras 3.* to having to use the keras.optimizers.get method and providing a class_name and config to deserialize. Relevant for lstm and feedforward autoencoders.
Model config now requires for tensorflow.keras.models.Sequential to define input_shape in its layers, otherwise model is not properly built after compiling and prior to fitting. Relevant for KerasRawModelRegressor.

KerasBaseEstimator underwent the most changes.

We now need to run the __init__ of the KerasRegressor with the expected kwargs for proper initialisation of the object, but the kwargs will always take precedence for fit, predict and compile, so they are mostly for making keras happy.
Saving model for pickling was changed ƒrom h5 format, which is now considered legacy to keras native zip format, and the file is now stored to a temporary file and read into a buffer instead of using the h5py library, since save_model and load_model now exclusively expect an actual file instead of an io buffer.
Current implementation of model preparation for fit depends on the __call__ method, which is no longer used in keras 3.*, and was changed to _prepare_model to replicate the same behaviour right before we call our own fit since it requires the X and Y to be present to set the n_features and n_features_out.
History is no longer returned from calling fit and must be extracted from under model.history.
Manually stored history now resides in self._history instead of self.history to avoid attribute name clash.

Adjustments to tests were made:

input_shape now resides under input.shape in model.layers.
optimizer_kwargs.lr is now optimizer_kwargs.learning_rate

To accommodate the upgrade, next changes to the code were made: * Since `keras 3.*` now clones the input `regressor` that is provided to the `BaseEstimator` into a `regressor_` attribute, and ONLY updates the `regressor_` one after fitting, metadata extraction in `gordo.builder.build_model.ModelBuilder` was changed to seek the `regressor_` in the `steps`. * Accessing `keras.optimizers` has changed in `keras 3.*` to having to use the `keras.optimizers.get` method and providing a `class_name` and `config` to deserialize. Relevant for lstm and feedforward autoencoders. * Model config now requires for `tensorflow.keras.models.Sequential` to define `input_shape` in its layers, otherwise model is not properly built after compiling and prior to fitting. Relevant for `KerasRawModelRegressor`. `KerasBaseEstimator` underwent the most changes. * We now need to run the `__init__` of the KerasRegressor with the expected `kwargs` for proper initialisation of the object, but the `kwargs` will always take precedence for `fit`, `predict` and `compile`, so they are mostly for making `keras` happy. * Saving model for pickling was changed ƒrom `h5` format, which is now considered legacy to `keras` native zip format, and the file is now stored to a temporary file and read into a buffer instead of using the `h5py` library, since `save_model` and `load_model` now exclusively expect an actual file instead of an io buffer. * Current implementation of model preparation for fit depends on the `__call__` method, which is no longer used in `keras 3.*`, and was changed to `_prepare_model` to replicate the same behaviour right before we call our own `fit` since it requires the `X` and `Y` to be present to set the `n_features` and `n_features_out`. * `History` is no longer returned from calling `fit` and must be extracted from under `model.history`. * Manually stored history now resides in `self._history` instead of `self.history` to avoid attribute name clash. Adjustments to tests were made: * `input_shape` now resides under `input.shape` in `model.layers`. * `optimizer_kwargs.lr` is now `optimizer_kwargs.learning_rate`

koropets · 2024-06-26T09:24:46Z

gordo/builder/build_model.py

+            if key.endswith(
+                "_"
+            ):  # keras3 clones the regressor into regressor_ and never updates original regressor


Suggested change

if key.endswith(

"_"

): # keras3 clones the regressor into regressor_ and never updates original regressor

if key == "regressor": # keras3 clones the regressor into regressor_ and never updates original regressor

continue

endwith is too broad to fix one back-compatibility issue

koropets · 2024-06-26T09:34:04Z

gordo/machine/model/models.py

+            *self._fit_kwargs,
+            *self._predict_kwargs,
+            *self._compile_kwargs,


Suggested change

*self._fit_kwargs,

*self._predict_kwargs,

*self._compile_kwargs,

*KerasRegressor._fit_kwargs,

*KerasRegressor._predict_kwargs,

*KerasRegressor._compile_kwargs,

in case of self not clear what an exact source of these constants

koropets · 2024-06-26T09:43:36Z

gordo/machine/model/models.py

@@ -301,25 +313,25 @@ def get_params(self, **params):
            Parameters used in this estimator
        """
        params = super().get_params(**params)
-        params.pop("build_fn", None)
+        params.pop("model", None)


Suggested change

params.pop("model", None)

unnecessary call

koropets · 2024-06-26T09:51:53Z

tests/gordo/machine/model/test_raw_keras.py

+                                    input_shape:
+                                        - 4


Suggested change

input_shape:

- 4

input_shape: [4]

just suggestion

koropets · 2024-06-26T09:52:11Z

tests/gordo/machine/model/test_raw_keras.py

+                                    input_shape:
+                                        - 1


Suggested change

input_shape:

- 1

…ressor` `layers`

hugobettmach · 2024-06-27T10:48:35Z

gordo/machine/model/factories/feedforward_autoencoder.py

@@ -88,8 +88,9 @@ class (e.x. Adam(lr=0.01,beta_1=0.9, beta_2=0.999)).  If no arguments are

    # Instantiate optimizer with kwargs
    if isinstance(optimizer, str):
-        Optim = getattr(keras.optimizers, optimizer)
-        optimizer = Optim(**optimizer_kwargs)


I think this previous approach still works though. It's what we have implemented in aion and runs fine. But I'm ok with changing it.

hugobettmach · 2024-06-27T10:49:34Z

gordo/machine/model/models.py

 import tensorflow.keras.models
 from tensorflow.keras.models import load_model, save_model
 from tensorflow.keras.preprocessing.sequence import pad_sequences, TimeseriesGenerator
-from tensorflow.keras.wrappers.scikit_learn import KerasRegressor as BaseWrapper
-from tensorflow.keras.callbacks import History
+from scikeras.wrappers import KerasRegressor


Interesting I wasn't aware about this. Good you found this out.

hugobettmach · 2024-06-27T10:52:50Z

gordo/machine/model/models.py

+            *KerasRegressor._predict_kwargs,
+            *KerasRegressor._compile_kwargs,
+        }
+        KerasRegressor.__init__(


This is indeed a bit strange, but oh well, as long as it works. Fortunately we won't need this wrapping business in aion.

hugobettmach

Wow quite a lot of tricky changes to make, good work @RollerKnobster in finding these out. It's hard to have an opinion if they will work or not but once we have a pre-release of gordo-anomaly I will use my local dieta to test it

RollerKnobster requested review from koropets, hugobettmach and Afonso-2403 June 26, 2024 08:45

RollerKnobster self-assigned this Jun 26, 2024

RollerKnobster added 2 commits June 26, 2024 11:50

Fix: formatting

6259b05

Fix: remove autodoc_mock_imports = ["tensorflow"] from docs conf.py

c53efbf

koropets reviewed Jun 26, 2024

View reviewed changes

RollerKnobster added 2 commits June 26, 2024 13:15

Fix: minor refactoring

29a715f

Fix: minor refactoring

9680e35

RollerKnobster requested a review from koropets June 26, 2024 10:43

RollerKnobster added 3 commits June 26, 2024 13:58

Fix: remove input_shape from subsequent layers in `KerasRawModelReg…

d6ab506

…ressor` `layers`

Fix: reformat KerasRawModelRegressor doc example

5733d98

Fix: reformat KerasRawModelRegressor doc example

7ecaed7

koropets approved these changes Jun 26, 2024

View reviewed changes

RollerKnobster merged commit b327587 into master Jun 26, 2024
17 checks passed

RollerKnobster deleted the feature/rivan/163526-upgrade-tensorflow-and-keras branch June 26, 2024 11:59

hugobettmach reviewed Jun 27, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Feat: Upgrade `tensorflow` to `2.16.1` and `keras` to `3.4.0` #1385

Feat: Upgrade `tensorflow` to `2.16.1` and `keras` to `3.4.0` #1385

RollerKnobster commented Jun 26, 2024 •

edited

Loading

koropets Jun 26, 2024 •

edited

Loading

koropets Jun 26, 2024 •

edited

Loading

koropets Jun 26, 2024

koropets Jun 26, 2024

koropets Jun 26, 2024

hugobettmach Jun 27, 2024

hugobettmach Jun 27, 2024

hugobettmach Jun 27, 2024

hugobettmach left a comment

Feat: Upgrade tensorflow to 2.16.1 and keras to 3.4.0 #1385

Feat: Upgrade tensorflow to 2.16.1 and keras to 3.4.0 #1385

Conversation

RollerKnobster commented Jun 26, 2024 • edited Loading

koropets Jun 26, 2024 • edited Loading

Choose a reason for hiding this comment

koropets Jun 26, 2024 • edited Loading

Choose a reason for hiding this comment

koropets Jun 26, 2024

Choose a reason for hiding this comment

koropets Jun 26, 2024

Choose a reason for hiding this comment

koropets Jun 26, 2024

Choose a reason for hiding this comment

hugobettmach Jun 27, 2024

Choose a reason for hiding this comment

hugobettmach Jun 27, 2024

Choose a reason for hiding this comment

hugobettmach Jun 27, 2024

Choose a reason for hiding this comment

hugobettmach left a comment

Choose a reason for hiding this comment

Feat: Upgrade `tensorflow` to `2.16.1` and `keras` to `3.4.0` #1385

Feat: Upgrade `tensorflow` to `2.16.1` and `keras` to `3.4.0` #1385

RollerKnobster commented Jun 26, 2024 •

edited

Loading

koropets Jun 26, 2024 •

edited

Loading

koropets Jun 26, 2024 •

edited

Loading