"""
Model conversion
=========================
Convert a Python model into C code
"""
from . import trees
from . import net
from . import bayes
from . import distance
from . import mixture
from . import neighbors
[docs]class Model():
"""Inference model powered by emlearn
Wrapper around the underlying C code.
Allows using the model in Python code, for evaluation/testing.
Follows scikit-learn API conventions.
"""
def __init__(self):
pass
[docs] def save(self, name : str, file : str = None) -> str:
"""Generate C code
:param name: Name for the model. Must be valid C identifier
:param file: Filepath for where to save the C code (optional)
:return: The generated C code
"""
[docs] def predict(self, X):
"""Run inference, return classes / regression values"""
pass
[docs] def predict_proba(self, X):
"""Run inference, return probabilities"""
pass
[docs] def score_samples(self, X):
"""Run inference, return anomaly/outlier scores"""
pass
[docs]def convert(estimator,
kind : str = None,
method: str = 'loadable',
dtype: str ='float',
return_type: str = 'classifier',
**kwargs,
) -> Model:
"""Convert model to C
:param kind: Explicit name for the type of model. Useful if the model is a subclass of a supported model class
:param method: The inference strategy to use. inline|loadable
:param dtype: Datatype to use for features. Can be used to enable quantization
:param return_type: Return type of the model. 'classifier' (default) creates a classifier (output binarized when needed), 'regressor' creates a regressor (output type is float).
:return: A Estimator like class, that uses C code for inference
"""
if kind is None:
kind = type(estimator).__name__
# Use name instead of instance to avoid hard dependency on the libraries
if kind in set(trees.SUPPORTED_ESTIMATORS):
# return_type is intentionally not passed through - the Wrapper will guess based on Class name
return trees.Wrapper(estimator, method, dtype=dtype, **kwargs)
elif kind in ['EllipticEnvelope']:
return distance.Wrapper(estimator, method, dtype=dtype)
elif kind in ['GaussianMixture', 'BayesianGaussianMixture']:
return mixture.Wrapper(estimator, method, dtype=dtype)
elif kind in ('MLPClassifier', 'MLPRegressor'):
return net.convert_sklearn_mlp(estimator, method, return_type=return_type, **kwargs)
elif kind == 'Sequential':
return net.convert_keras(estimator, method, return_type=return_type, **kwargs)
elif kind == 'GaussianNB':
return bayes.Wrapper(estimator, method)
elif kind in ['KNeighborsClassifier']:
return neighbors.convert_sklearn(estimator, inference=method)
else:
raise ValueError("Unknown model type: '{}'".format(kind))