-
Notifications
You must be signed in to change notification settings - Fork 126
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge pull request #1171 from yimuchen/ml_tools_tf
feat: tensorflow wrapper
- Loading branch information
Showing
6 changed files
with
238 additions
and
19 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,113 @@ | ||
| import warnings | ||
|
|
||
| import numpy | ||
|
|
||
| _tf_import_error = None | ||
| try: | ||
| import tensorflow | ||
| except (ImportError, ModuleNotFoundError) as err: | ||
| _tf_import_error = err | ||
|
|
||
| from .helper import nonserializable_attribute, numpy_call_wrapper | ||
|
|
||
|
|
||
| class tf_wrapper(nonserializable_attribute, numpy_call_wrapper): | ||
| """ | ||
| Wrapper for running tensorflow inference with awkward/dask-awkward inputs. | ||
| """ | ||
|
|
||
| def __init__(self, tf_model: str): | ||
| """ | ||
| As models are not guaranteed to be directly serializable, the use will | ||
| need to pass the model as files saved using the `tf.keras.save` method | ||
| [1]. If the user is attempting to run on the clusters, the model file | ||
| will need to be passed to the worker nodes in a way which preserves the | ||
| file path. | ||
| [1] | ||
| https://www.tensorflow.org/guide/keras/serialization_and_saving#saving | ||
| Parameters ---------- | ||
| - tf_model: Path to the tensorflow model file to load | ||
| """ | ||
| if _tf_import_error is not None: | ||
| warnings.warn( | ||
| "Users should make sure the torch package is installed before proceeding!\n" | ||
| "> pip install tensorflow\n" | ||
| "or\n" | ||
| "> conda install tensorflow", | ||
| UserWarning, | ||
| ) | ||
| raise _tf_import_error | ||
|
|
||
| nonserializable_attribute.__init__(self, ["model"]) | ||
| self.tf_model = tf_model | ||
|
|
||
| def _create_model(self): | ||
| """ | ||
| Loading in the model from the model file. We simply rely on Tensorflow | ||
| to automatically load the accelerator resources. | ||
| # TODO: More control over accelerator resources? | ||
| """ | ||
| return tensorflow.keras.models.load_model(self.tf_model) | ||
|
|
||
| def validate_numpy_input(self, *args: numpy.array, **kwargs: numpy.array) -> None: | ||
| """ | ||
| Here we are assuming that the model contains the required information | ||
| for parsing the input numpy array(s), and that the input numpy array(s) | ||
| is the first argument of the user method call. | ||
| """ | ||
| model_input = self.model.input_shape | ||
| input_arr = args[0] # Getting the input array | ||
|
|
||
| def _equal_shape(mod_in: tuple, arr_shape: tuple) -> None: | ||
| """Tuple of input shape and array shape""" | ||
| assert len(mod_in) == len( | ||
| arr_shape | ||
| ), f"Mismatch number of axis (model: {mod_in}; received: {arr_shape})" | ||
| match = [ | ||
| (m == a if m is not None else True) for m, a in zip(mod_in, arr_shape) | ||
| ] | ||
| assert numpy.all( | ||
| match | ||
| ), f"Mismatch shape (model: {mod_in}; received: {arr_shape})" | ||
|
|
||
| if isinstance(model_input, tuple): | ||
| # Single input model | ||
| _equal_shape(model_input, input_arr.shape) | ||
| else: | ||
| assert len(input_arr) == len( | ||
| model_input | ||
| ), f"Mismatch number of inputs (model: {len(model_input)}; received: {len(input_arr)})" | ||
| for model_shape, arr in zip(model_input, input_arr): | ||
| _equal_shape(model_shape, arr.shape) | ||
|
|
||
| def numpy_call(self, *args: numpy.array, **kwargs: numpy.array) -> numpy.array: | ||
| """ | ||
| Evaluating the numpy inputs via the `model.__call__` method. With a | ||
| trivial conversion for tensors for the numpy inputs. | ||
| TODO: Do we need to evaluate using `predict` [1]? Since array batching | ||
| is already handled by dask. | ||
| [1] | ||
| https://keras.io/getting_started/faq/#whats-the-difference-between-model-methods-predict-and-call | ||
| """ | ||
| args = [ | ||
| ( | ||
| tensorflow.convert_to_tensor(arr) | ||
| if arr.flags["WRITEABLE"] | ||
| else tensorflow.convert_to_tensor(numpy.copy(arr)) | ||
| ) | ||
| for arr in args | ||
| ] | ||
| kwargs = { | ||
| key: ( | ||
| tensorflow.convert_to_tensor(arr) | ||
| if arr.flags["WRITABLE"] | ||
| else tensorflow.convert_to_tensor(numpy.copy(arr)) | ||
| ) | ||
| for key, arr in kwargs.items() | ||
| } | ||
| return self.model(*args, **kwargs).numpy() |
Binary file not shown.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters