class ValidateTestSize(validators.Validator):
def __init__(self, **kwargs):
super(ValidateTestSize, self).__init__()
int_args = {
'minimum': kwargs.get('int_minimum'),
'maximum': kwargs.get('int_maximum'),
}
float_args = {
'minimum': kwargs.get('float_minimum'),
'maximum': kwargs.get('float_maximum'),
}
self.validate_int = validators.Integer(**int_args)
self.validate_float = ValidateFloat(**float_args)
def __call__(self, value):
import numbers
if value is not None and not isinstance(value, numbers.Number):
try:
test = int(value)
self.validate_int.check_range(test)
return test
except ValueError:
value = float(value)
self.validate_float.check_range(value)
return value
def format(self, value):
return str(value)
def __init__(self, **kwargs):
super(ValidateTestSize, self).__init__()
int_args = {
'minimum': kwargs.get('int_minimum'),
'maximum': kwargs.get('int_maximum'),
}
float_args = {
'minimum': kwargs.get('float_minimum'),
'maximum': kwargs.get('float_maximum'),
}
self.validate_int = validators.Integer(**int_args)
self.validate_float = ValidateFloat(**float_args)
class ValidateTestSize(validators.Validator):
def __init__(self, **kwargs):
super(ValidateTestSize, self).__init__()
int_args = {"minimum": kwargs.get("int_minimum"), "maximum": kwargs.get("int_maximum")}
float_args = {"minimum": kwargs.get("float_minimum"), "maximum": kwargs.get("float_maximum")}
self.validate_int = validators.Integer(**int_args)
self.validate_float = ValidateFloat(**float_args)
def __call__(self, value):
import numbers
if value is not None and not isinstance(value, numbers.Number):
try:
test = int(value)
self.validate_int.check_range(test)
return test
except ValueError:
value = float(value)
self.validate_float.check_range(value)
return value
def format(self, value):
return str(value)
def __init__(self, **kwargs):
super(ValidateTestSize, self).__init__()
int_args = {"minimum": kwargs.get("int_minimum"), "maximum": kwargs.get("int_maximum")}
float_args = {"minimum": kwargs.get("float_minimum"), "maximum": kwargs.get("float_maximum")}
self.validate_int = validators.Integer(**int_args)
self.validate_float = ValidateFloat(**float_args)
class MCTrain(OptionRemoteReportingCommand):
model = Option(require=True,
validate=ValidateLocalFile(extension="pkl",
subdir='classifiers',
nohandle=True))
target = Option(require=True, validate=validators.Fieldname())
reset = Option(require=False, default=False, validate=validators.Boolean())
textmodel = Option(require=False, default='hashing')
test_size = Option(require=False,
default=0.25,
validate=ValidateTestSize(int_minimum=0,
float_minimum=0,
float_maximum=1))
_C = Option(require=False,
default=1.0,
validate=ValidateFloat(minimum=0.0))
reduce_code = """
import os, sys, itertools, collections, numbers
import numpy as np
import scipy.sparse as sp
from multiclassify import process_records
# Can try one of these 2
from gensim.models import LsiModel, TfidfModel, LdaModel
from sklearn.linear_model import LogisticRegression
# from sklearn.preprocessing import MultiLabelBinarizer
from sklearn.preprocessing import LabelEncoder
from sklearn.cross_validation import train_test_split
from sklearn.externals import joblib
if __name__ == "__channelexec__":
args = channel.receive()
fields = args.get('fieldnames') or ['_raw']
target = args['target']
records = []
for record in channel:
if not record:
break
records.append(record)
def is_number(str):
try:
n = float(str)
return True
except ValueError:
return False
if records:
records = np.array(records)
# Try loading existing model
encoder = None
est = None
if not args['reset']:
try:
model = joblib.load(args['model'])
encoder = model['encoder']
est = model['est']
est.densify()
except:
encoder = None
est = None
X, y_labels, textmodel = process_records(records, fields, target, args['textmodel'])
if not encoder:
encoder = LabelEncoder()
y = encoder.fit_transform(y_labels)
print >> sys.stderr, "CLASSES:", encoder.classes_
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=args['test_size'])
if not est:
kwargs = {
'C': args['_C'],
}
if len(encoder.classes_) <= 2:
kwargs['multi_class'] = 'ovr'
else:
kwargs['multi_class'] = 'multinomial'
kwargs['solver'] = 'lbfgs'
est = LogisticRegression(**kwargs)
est.fit(X_train, y_train)
score = est.score(X_test, y_test) if len(y_test) > 0 else None
# Reduce footprint on disk
if (est.coef_ == 0).sum() > (est.coef_.size / 2):
est.sparsify()
model = {
'encoder': encoder,
'est': est,
'score': score,
'X': X,
'y_labels': y_labels,
'target': args['target'],
'fields': fields,
}
if textmodel:
model['text'] = args['textmodel']
textmodel.save(args['model'].replace(".pkl",".%s" % args['textmodel']))
elif args['textmodel'] == 'hashing':
model['text'] = args['textmodel']
joblib.dump(model, args['model'])
print >> sys.stderr, "END"
channel.send({
'model': args['model'],
'score': score.item() if score else None,
'training_size': X_train.shape[0],
'test_size': X_test.shape[0]
})
"""
def __dir__(self):
return ['reset', 'model', 'textmodel', 'test_size', 'target', '_C']
@Configuration(clear_required_fields=True)
def map(self, records):
try:
for record in records:
yield record
except:
yield {}
