def __init__(self, parameters, cross_experiment_key, **kwargs):
"""A KeyMaker class dedicated to creating hyperparameter keys, which determine when
experiments were executed using identical hyperparameters. Two separate instances of
:class:`experiments.CVExperiment` should produce identical `hyperparameter_key` s if their
hyperparameters are the same (or close enough)
Parameters
----------
parameters: Dict
All the parameters to be included when creating the key hash. Keys should correspond to
parameter names, and values should be the values of the corresponding keys
cross_experiment_key: Str
The key produced by the active Environment via
:class:`key_handler.CrossExperimentKeyMaker`, used for determining when a
hyperparameter key has already been tested under the same cross-experiment parameters
**kwargs: Dict
Additional arguments supplied to :meth:`key_handler.KeyMaker.__init__`"""
self.cross_experiment_key = cross_experiment_key
self.is_task_keras = (
hasattr(G.Env, "current_task")
and G.Env.current_task
and G.Env.current_task.module_name == "keras"
)
if self.is_task_keras:
parameters = deepcopy(parameters)
#################### Initialize and Parameterize Dummy Model ####################
temp_model = initialize_dummy_model(
parameters["model_initializer"],
parameters["model_init_params"]["build_fn"],
parameters["model_extra_params"],
)
temp_layers, temp_compile_params = parameterize_compiled_keras_model(temp_model)
#################### Process Parameters ####################
# noinspection PyUnusedLocal
def _visit(path, key, value):
"""If `key` not in ('input_shape', 'input_dim'), return True. Else, return False"""
return key not in ("input_shape", "input_dim")
temp_layers = remap(temp_layers, visit=_visit)
parameters["model_init_params"]["layers"] = temp_layers
parameters["model_init_params"]["compile_params"] = temp_compile_params
if "params" in parameters["model_extra_params"]:
parameters["model_extra_params"] = {
_k: _v for _k, _v in parameters["model_extra_params"].items() if _k != "params"
}
KeyMaker.__init__(self, parameters, **kwargs)
def test_parameterize_compiled_keras_model(model, layers, compile_params):
if get_distribution("keras").version < "2.2.0":
layers.insert(
0,
{
"class_name": "InputLayer",
"__hh_default_args": None,
"__hh_default_kwargs": None,
"__hh_used_args": None,
"__hh_used_kwargs": None,
},
)
assert parameterize_compiled_keras_model(model()) == (layers,
compile_params)
def keras_prep_workflow(model_initializer, build_fn, extra_params,
source_script):
"""Conduct preparation steps necessary before hyperparameter optimization on a `Keras` model.
Such steps include parsing and modifying `build_fn` to be of the form used by
:class:`hyperparameter_hunter.optimization_core.BaseOptimizationProtocol`, compiling a dummy
model to identify universal locations of given hyperparameter choices, and creating a simplified
characterization of the models to be built during optimization in order to enable collection of
similar Experiments
Parameters
----------
model_initializer: :class:`keras.wrappers.scikit_learn.<KerasClassifier; KerasRegressor>`
A descendant of :class:`keras.wrappers.scikit_learn.BaseWrapper` used to build a Keras model
build_fn: Callable
The `build_fn` value provided to :meth:`keras.wrappers.scikit_learn.BaseWrapper.__init__`.
Expected to return a compiled Keras model. May contain hyperparameter space choices
extra_params: Dict
The parameters expected to be passed to the extra methods of the compiled Keras model. Such
methods include (but are not limited to) `fit`, `predict`, and `predict_proba`. Some of the
common parameters given here include `epochs`, `batch_size`, and `callbacks`
source_script: Str
Absolute path to a Python file. Expected to end with one of the following extensions:
'.py', '.ipynb'
Returns
-------
reusable_build_fn: Callable
Modified `build_fn` in which hyperparameter space choices are replaced by dict lookups, and
the signature is given a standard name, and additional input parameters necessary for reuse
reusable_wrapper_params: Dict
The parameters expected to be passed to the extra methods of the compiled Keras model. Such
methods include (but are not limited to) `fit`, `predict`, and `predict_proba`. Some of the
common parameters given here include `epochs`, `batch_size`, and `callbacks`
dummy_layers: List
The layers of a compiled dummy Keras model constructed according to the given
hyperparameters, in which each layer is a dict containing at least the following: the name
of the layer class, allowed and used args, and default and used kwargs
dummy_compile_params: Dict
The parameters used on the `compile` call for the dummy model. If a parameter is accepted
by the `compile` method, but is not explicitly given, its default value is included in
`dummy_compile_params`"""
#################### Prepare Model-Builder String ####################
temp_builder_name = "__temp_model_builder"
reusable_build_fn, expected_params = rewrite_model_builder(
stringify_model_builder(build_fn))
temp_model_file_str = build_temp_model_file(reusable_build_fn,
source_script)
#################### Save and Import Temporary Model Builder ####################
write_python_source(
temp_model_file_str,
"{}/{}.py".format(os.path.split(__file__)[0], temp_builder_name))
if temp_builder_name in sys.modules:
del sys.modules[temp_builder_name]
try:
from .__temp_model_builder import build_fn as temp_build_fn
except:
raise
#################### Translate Hyperparameter Names to Universal Paths ####################
wrapper_params = dict(
params={_k: eval(_v)
for _k, _v in expected_params.items()},
**extra_params)
wrapper_params, dummified_params = check_dummy_params(wrapper_params)
if ("optimizer_params" in dummified_params) and ("optimizer"
in dummified_params):
raise ValueError(
"Unable to optimize both `optimizer` and `optimizer_params`. Please try optimizing them separately"
)
compiled_dummy = initialize_dummy_model(model_initializer, temp_build_fn,
wrapper_params)
dummy_layers, dummy_compile_params = parameterize_compiled_keras_model(
compiled_dummy)
merged_compile_params = merge_compile_params(dummy_compile_params,
dummified_params)
# FLAG: Will need to deal with capitalization conflicts when comparing similar experiments: `optimizer`='Adam' vs 'adam'
consolidated_layers = consolidate_layers(dummy_layers,
class_name_key=False,
separate_args=False)
wrapper_params = deep_restricted_update(wrapper_params, dummified_params)
return (temp_build_fn, wrapper_params, consolidated_layers,
merged_compile_params)
def test_parameterize_compiled_keras_model(model, layers, compile_params):
assert parameterize_compiled_keras_model(model()) == (layers, compile_params)
def keras_prep_workflow(model_initializer, build_fn, extra_params,
source_script):
"""Conduct preparation steps necessary before hyperparameter optimization on a `Keras` model.
Such steps include parsing and modifying `build_fn` to be of the form used by
:class:`hyperparameter_hunter.optimization_core.BaseOptimizationProtocol`, compiling a dummy
model to identify universal locations of given hyperparameter choices, and creating a simplified
characterization of the models to be built during optimization in order to enable collection of
similar Experiments
Parameters
----------
model_initializer: :class:`keras.wrappers.scikit_learn.<KerasClassifier; KerasRegressor>`
A descendant of :class:`keras.wrappers.scikit_learn.BaseWrapper` used to build a Keras model
build_fn: Callable
The `build_fn` value provided to :meth:`keras.wrappers.scikit_learn.BaseWrapper.__init__`.
Expected to return a compiled Keras model. May contain hyperparameter space choices
extra_params: Dict
The parameters expected to be passed to the extra methods of the compiled Keras model. Such
methods include (but are not limited to) `fit`, `predict`, and `predict_proba`. Some of the
common parameters given here include `epochs`, `batch_size`, and `callbacks`
source_script: Str
Absolute path to a Python file. Should end with one of following extensions: ".py", ".ipynb"
Returns
-------
reusable_build_fn: Callable
Modified `build_fn` in which hyperparameter space choices are replaced by dict lookups, and
the signature is given a standard name, and additional input parameters necessary for reuse
reusable_wrapper_params: Dict
The parameters expected to be passed to the extra methods of the compiled Keras model. Such
methods include (but are not limited to) `fit`, `predict`, and `predict_proba`. Some of the
common parameters given here include `epochs`, `batch_size`, and `callbacks`
dummy_layers: List
The layers of a compiled dummy Keras model constructed according to the given
hyperparameters, in which each layer is a dict containing at least the following: the name
of the layer class, allowed and used args, and default and used kwargs
dummy_compile_params: Dict
The parameters used on the `compile` call for the dummy model. If a parameter is accepted
by the `compile` method, but is not explicitly given, its default value is included in
`dummy_compile_params`"""
#################### Set Temporary Model-Builder Module Location ####################
temp_module_name = f"__temp_model_builder_{datetime.now().strftime('%Y-%m-%d_%H-%M-%S-%f')}"
temp_module_dot_path = f"{TEMP_MODULES_DOT_PATH}.{temp_module_name}"
temp_module_filepath = f"{TEMP_MODULES_DIR_PATH}/{temp_module_name}.py"
#################### Prepare Model-Builder String ####################
reusable_build_fn, expected_params = rewrite_model_builder(
stringify_model_builder(build_fn))
temp_module_str = build_temp_model_file(reusable_build_fn, source_script)
#################### Save and Import Temporary Model Builder ####################
write_python(temp_module_str, temp_module_filepath)
if temp_module_name in sys.modules:
del sys.modules[temp_module_name]
temp_module_spec = spec_from_file_location(temp_module_dot_path,
temp_module_filepath)
temp_module = module_from_spec(temp_module_spec)
temp_module_spec.loader.exec_module(temp_module)
temp_build_fn = temp_module.build_fn
#################### Translate Hyperparameter Names to Universal Paths ####################
wrapper_params = dict(
params={k: eval(v)
for k, v in expected_params.items()},
**extra_params)
# TODO: Intercept space choices that use callables (like `Categorical([glorot_normal(), orthogonal()])`)
# TODO: Can't deal with them yet, due to imports unavailable in this context. Raise exception
wrapper_params, dummified_params = check_dummy_params(wrapper_params)
if ("optimizer_params" in dummified_params) and ("optimizer"
in dummified_params):
raise ValueError(
"Can't optimize `optimizer` with `optimizer_params`. Try them separately"
)
compiled_dummy = initialize_dummy_model(model_initializer, temp_build_fn,
wrapper_params)
dummy_layers, dummy_compile_params = parameterize_compiled_keras_model(
compiled_dummy)
merged_compile_params = merge_compile_params(dummy_compile_params,
dummified_params)
# FLAG: Will need to deal with capitalization conflicts when comparing similar experiments: `optimizer`="Adam" vs "adam"
consolidated_layers = consolidate_layers(dummy_layers,
class_name_key=False,
split_args=False)
wrapper_params = deep_restricted_update(wrapper_params, dummified_params)
return (temp_build_fn, wrapper_params, consolidated_layers,
merged_compile_params)