def compare_penalties(fileData):
params = utilities.Parameters(lowerCaseFlag=False,
removeStopWordsFlag=False,
stemFlag=False,
maxFeatures=7363,
ngramRange=(1, 1),
tfidfFlags=(False, False))
svmL1 = svm.LinearSVC(penalty='l1', dual=False)
svmL2 = svm.LinearSVC(penalty='l2')
Corpus, pipelineL1 = utilities.getInfoFromParameters(
fileData, params, svmL1)
Corpus, pipelineL2 = utilities.getInfoFromParameters(
fileData, params, svmL2)
splits = 10
outer_cv = StratifiedKFold(n_splits=splits, shuffle=True, random_state=42)
# CV for L1 estimator and L2 estimator which returns the F1-score and Accuracy.
scoring = ['accuracy', 'f1_macro']
scoresL2 = cross_validate(pipelineL2,
X=Corpus[preprocessing.COMMENT],
y=Corpus[preprocessing.CLASS],
scoring=scoring,
cv=outer_cv)
scoresL1 = cross_validate(pipelineL1,
X=Corpus[preprocessing.COMMENT],
y=Corpus[preprocessing.CLASS],
scoring=scoring,
cv=outer_cv)
for i in range(1, splits):
scoresL1['test_accuracy'][0] += scoresL1['test_accuracy'][i]
scoresL1['test_f1_macro'][0] += scoresL1['test_f1_macro'][i]
scoresL2['test_accuracy'][0] += scoresL2['test_accuracy'][i]
scoresL2['test_f1_macro'][0] += scoresL2['test_f1_macro'][i]
scoresL1['test_accuracy'][0] /= splits
scoresL1['test_f1_macro'][0] /= splits
scoresL2['test_accuracy'][0] /= splits
scoresL2['test_f1_macro'][0] /= splits
print("L1 accuracy: ", scoresL1['test_accuracy'][0], " - L2 accuracy: ",
scoresL2['test_accuracy'][0])
print("L1 F1: ", scoresL1['test_f1_macro'][0], " - L2 F1: ",
scoresL2['test_f1_macro'][0])
def compare_penalties(fileData):
params = utilities.Parameters(lowerCaseFlag=False,
removeStopWordsFlag=False,
stemFlag=False,
maxFeatures=7363,
ngramRange=(1, 1),
tfidfFlags=(False, False))
lrL1 = LogisticRegression(penalty='l1', solver='saga', tol=0.01)
lrL2 = LogisticRegression(penalty='l2')
Corpus, pipelineL1 = utilities.getInfoFromParameters(
fileData, params, lrL1)
Corpus, pipelineL2 = utilities.getInfoFromParameters(
fileData, params, lrL2)
splits = 2
outer_cv = StratifiedKFold(n_splits=splits, shuffle=True, random_state=42)
# CV for L1 estimator and L2 estimator which returns f1 and accuracy and we will compare it
scoring = ['accuracy', 'f1_macro']
scoresL1 = cross_validate(pipelineL1,
X=Corpus[preprocessing.COMMENT],
y=Corpus[preprocessing.CLASS],
scoring=scoring,
cv=outer_cv)
scoresL2 = cross_validate(pipelineL2,
X=Corpus[preprocessing.COMMENT],
y=Corpus[preprocessing.CLASS],
scoring=scoring,
cv=outer_cv)
for i in range(1, splits):
scoresL1['test_accuracy'][0] += scoresL1['test_accuracy'][i]
scoresL1['test_f1_macro'][0] += scoresL1['test_f1_macro'][i]
scoresL2['test_accuracy'][0] += scoresL2['test_accuracy'][i]
scoresL2['test_f1_macro'][0] += scoresL2['test_f1_macro'][i]
scoresL1['test_accuracy'][0] /= splits
scoresL1['test_f1_macro'][0] /= splits
scoresL2['test_accuracy'][0] /= splits
scoresL2['test_f1_macro'][0] /= splits
print("L1 accuracy: ", scoresL1['test_accuracy'][0], " - L2 accuracy: ",
scoresL2['test_accuracy'][0])
print("L1 F1: ", scoresL1['test_f1_macro'][0], " - L2 F1: ",
scoresL2['test_f1_macro'][0])
if __name__ == "__main__":
# Construct parameters.
parametersList = list()
for lowerCaseFlag in [False, True]:
for removeStopWordsFlag in [False, True]:
for stemFlag in [False, True]:
for maxFeatures in [1000, 5000, 7363]:
for ngramRange in [(1, 1), (1, 2), (1, 3)]:
for tfidfFlags in [(False, False), (True, False),
(False, True)]:
for alpha_value in [1, 0.001, 0.00001]:
parametersList.append(
utilities.Parameters(
lowerCaseFlag, removeStopWordsFlag,
stemFlag, maxFeatures, ngramRange,
tfidfFlags, alpha_value))
cnt = 0
# Go through all of the input files and configurations and export the results to a .csv file.
for input_file, output_file_path, singleFunctionalClass in [
("../input.txt", "output/outputNBdirectAlphaAll.csv", False),
("../input-functional.txt",
"output/outputNBdirectAlphaFunctionalAll.csv", True)
]:
with open(output_file_path, 'w') as output:
print(utilities.getHeader(singleFunctionalClass), file=output)
output.flush()
fileData = preprocessing.read_file(input_file)
# Construct parameters.
parametersList = list()
for lowerCaseFlag in [False, True]:
for removeStopWordsFlag in [False, True]:
for stemFlag in [False, True]:
for maxFeatures in [1000, 5000, 7363]:
for ngramRange in [(1, 1), (1, 2), (1, 3)]:
for alpha in [0.00001, 0.001, 1]:
for binarize in [0.0]:
for tfidfFlags in [(False, False)]:
parametersList.append(utilities.Parameters(
lowerCaseFlag,
removeStopWordsFlag,
stemFlag,
maxFeatures,
ngramRange,
tfidfFlags,
alpha,
binarize)
)
print("ParamsList created.\n")
count_file = 0
for input_file, output_file, is_functional in [("../input-functional.txt", "output-functional.csv", True), ("../input.txt", "output.csv", False)]:
with open(output_file, 'w') as output_file_print_target:
print("Using ",input_file, ", stats will be in ", output_file)
fileData = preprocessing.read_file(input_file)
# Print header in output file.
header = utilities.getHeader(is_functional)