1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345
|
# -*- coding: utf-8 -*-
"""This file is part of the TPOT library.
TPOT was primarily developed at the University of Pennsylvania by:
- Randal S. Olson (rso@randalolson.com)
- Weixuan Fu (weixuanf@upenn.edu)
- Daniel Angell (dpa34@drexel.edu)
- and many more generous open source contributors
TPOT is free software: you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as
published by the Free Software Foundation, either version 3 of
the License, or (at your option) any later version.
TPOT is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with TPOT. If not, see <http://www.gnu.org/licenses/>.
"""
import numpy as np
from sklearn.base import BaseEstimator, is_classifier, is_regressor
from sklearn.gaussian_process.kernels import Kernel
import inspect
class Operator(object):
"""Base class for operators in TPOT."""
root = False # Whether this operator type can be the root of the tree
import_hash = None
sklearn_class = None
arg_types = None
class ARGType(object):
"""Base class for parameter specifications."""
pass
def source_decode(sourcecode, verbose=0):
"""Decode operator source and import operator class.
Parameters
----------
sourcecode: string
a string of operator source (e.g 'sklearn.feature_selection.RFE')
verbose: int, optional (default: 0)
How much information TPOT communicates while it's running.
0 = none, 1 = minimal, 2 = high, 3 = all.
if verbose > 2 then ImportError will rasie during initialization
Returns
-------
import_str: string
a string of operator class source (e.g. 'sklearn.feature_selection')
op_str: string
a string of operator class (e.g. 'RFE')
op_obj: object
operator class (e.g. RFE)
"""
tmp_path = sourcecode.split(".")
op_str = tmp_path.pop()
import_str = ".".join(tmp_path)
try:
if sourcecode.startswith("tpot."):
exec("from {} import {}".format(import_str[4:], op_str))
else:
exec("from {} import {}".format(import_str, op_str))
op_obj = eval(op_str)
except Exception as e:
if verbose > 2:
raise ImportError("Error: could not import {}.\n{}".format(sourcecode, e))
else:
print(
"Warning: {} is not available and will not be used by TPOT.".format(
sourcecode
)
)
op_obj = None
return import_str, op_str, op_obj
def set_sample_weight(pipeline_steps, sample_weight=None):
"""Recursively iterates through all objects in the pipeline and sets sample weight.
Parameters
----------
pipeline_steps: array-like
List of (str, obj) tuples from a scikit-learn pipeline or related object
sample_weight: array-like
List of sample weight
Returns
-------
sample_weight_dict:
A dictionary of sample_weight
"""
sample_weight_dict = {}
if not isinstance(sample_weight, type(None)):
for (pname, obj) in pipeline_steps:
if inspect.getargspec(obj.fit).args.count("sample_weight"):
step_sw = pname + "__sample_weight"
sample_weight_dict[step_sw] = sample_weight
if sample_weight_dict:
return sample_weight_dict
else:
return None
def _is_selector(estimator):
selector_attributes = [
"get_support",
"transform",
"inverse_transform",
"fit_transform",
]
return all(hasattr(estimator, attr) for attr in selector_attributes)
def _is_transformer(estimator):
return hasattr(estimator, "fit_transform")
def _is_resampler(estimator):
return hasattr(estimator, "fit_resample")
def ARGTypeClassFactory(classname, prange, BaseClass=ARGType):
"""Dynamically create parameter type class.
Parameters
----------
classname: string
parameter name in a operator
prange: list
list of values for the parameter in a operator
BaseClass: Class
inherited BaseClass for parameter
Returns
-------
Class
parameter class
"""
return type(classname, (BaseClass,), {"values": prange})
def TPOTOperatorClassFactory(
opsourse, opdict, BaseClass=Operator, ArgBaseClass=ARGType, verbose=0
):
"""Dynamically create operator class.
Parameters
----------
opsourse: string
operator source in config dictionary (key)
opdict: dictionary
operator params in config dictionary (value)
regression: bool
True if it can be used in TPOTRegressor
classification: bool
True if it can be used in TPOTClassifier
BaseClass: Class
inherited BaseClass for operator
ArgBaseClass: Class
inherited BaseClass for parameter
verbose: int, optional (default: 0)
How much information TPOT communicates while it's running.
0 = none, 1 = minimal, 2 = high, 3 = all.
if verbose > 2 then ImportError will rasie during initialization
Returns
-------
op_class: Class
a new class for a operator
arg_types: list
a list of parameter class
"""
class_profile = {}
dep_op_list = {} # list of nested estimator/callable function
dep_op_type = {} # type of nested estimator/callable function
import_str, op_str, op_obj = source_decode(opsourse, verbose=verbose)
if not op_obj:
return None, None
else:
# define if the operator can be the root of a pipeline
if is_classifier(op_obj):
class_profile["root"] = True
optype = "Classifier"
elif is_regressor(op_obj):
class_profile["root"] = True
optype = "Regressor"
elif _is_selector(op_obj):
optype = "Selector"
elif _is_transformer(op_obj):
optype = "Transformer"
elif _is_resampler(op_obj):
optype = "Resampler"
else:
raise ValueError(
"optype must be one of: Classifier, Regressor, Selector, Transformer, or Resampler"
)
@classmethod
def op_type(cls):
"""Return the operator type.
Possible values:
"Classifier", "Regressor", "Selector", "Transformer"
"""
return optype
class_profile["type"] = op_type
class_profile["sklearn_class"] = op_obj
import_hash = {}
import_hash[import_str] = [op_str]
arg_types = []
for pname in sorted(opdict.keys()):
prange = opdict[pname]
if not isinstance(prange, dict):
classname = "{}__{}".format(op_str, pname)
arg_types.append(ARGTypeClassFactory(classname, prange, ArgBaseClass))
else:
for dkey, dval in prange.items():
dep_import_str, dep_op_str, dep_op_obj = source_decode(
dkey, verbose=verbose
)
if dep_import_str in import_hash:
import_hash[dep_import_str].append(dep_op_str)
else:
import_hash[dep_import_str] = [dep_op_str]
dep_op_list[pname] = dep_op_str
dep_op_type[pname] = dep_op_obj
if dval:
for dpname in sorted(dval.keys()):
dprange = dval[dpname]
classname = "{}__{}__{}".format(op_str, dep_op_str, dpname)
arg_types.append(
ARGTypeClassFactory(classname, dprange, ArgBaseClass)
)
class_profile["arg_types"] = tuple(arg_types)
class_profile["import_hash"] = import_hash
class_profile["dep_op_list"] = dep_op_list
class_profile["dep_op_type"] = dep_op_type
@classmethod
def parameter_types(cls):
"""Return the argument and return types of an operator.
Parameters
----------
None
Returns
-------
parameter_types: tuple
Tuple of the DEAP parameter types and the DEAP return type for the
operator
"""
return ([np.ndarray] + arg_types, np.ndarray) # (input types, return types)
class_profile["parameter_types"] = parameter_types
@classmethod
def export(cls, *args):
"""Represent the operator as a string so that it can be exported to a file.
Parameters
----------
args
Arbitrary arguments to be passed to the operator
Returns
-------
export_string: str
String representation of the sklearn class with its parameters in
the format:
SklearnClassName(param1="val1", param2=val2)
"""
op_arguments = []
if dep_op_list:
dep_op_arguments = {}
for dep_op_str in dep_op_list.values():
dep_op_arguments[dep_op_str] = []
for arg_class, arg_value in zip(arg_types, args):
aname_split = arg_class.__name__.split("__")
if isinstance(arg_value, str):
arg_value = '"{}"'.format(arg_value)
if len(aname_split) == 2: # simple parameter
op_arguments.append("{}={}".format(aname_split[-1], arg_value))
# Parameter of internal operator as a parameter in the
# operator, usually in Selector
else:
dep_op_arguments[aname_split[1]].append(
"{}={}".format(aname_split[-1], arg_value)
)
tmp_op_args = []
if dep_op_list:
# To make sure the inital operators is the first parameter just
# for better persentation
for dep_op_pname, dep_op_str in dep_op_list.items():
arg_value = dep_op_str # a callable function, e.g scoring function
doptype = dep_op_type[dep_op_pname]
if inspect.isclass(doptype): # a estimator
if (
issubclass(doptype, BaseEstimator)
or is_classifier(doptype)
or is_regressor(doptype)
or _is_transformer(doptype)
or _is_resampler(doptype)
or issubclass(doptype, Kernel)
):
arg_value = "{}({})".format(
dep_op_str, ", ".join(dep_op_arguments[dep_op_str])
)
tmp_op_args.append("{}={}".format(dep_op_pname, arg_value))
op_arguments = tmp_op_args + op_arguments
return "{}({})".format(op_obj.__name__, ", ".join(op_arguments))
class_profile["export"] = export
op_classname = "TPOT_{}".format(op_str)
op_class = type(op_classname, (BaseClass,), class_profile)
op_class.__name__ = op_str
return op_class, arg_types
|