class LogReg:
def __init__(self):
self.load_data()
self.clf = LogisticRegression(class_weight = 'balanced')
self.train()
self.predict()
def load_data(self):
train_csv = './data/train.csv'
test_csv = './data/test.csv'
df_train = pd.read_csv(train_csv, header=0)
df_test = pd.read_csv(test_csv, header=0)
arr_train = df_train.values
arr_test = df_test.values
self.train_X = arr_train[0::,1::]
self.train_Y = arr_train[0::, 0]
self.test_X = arr_test[0::, 1::]
self.test_ID = arr_test[0::,0]
def train(self):
self.clf.fit(self.train_X, self.train_Y)
def predict(self):
self.test_Y = self.clf.predict_proba(self.test_X)
def get_training_accuracy(self):
return (self.clf.score(self.train_X, self.train_Y))
def store_result(self):
df_out = pd.DataFrame()
df_out['Id'] = self.test_ID
df_out['Action'] = self.test_Y[0::,1]
df_out.to_csv('./data/results/c1_result.csv',index=False)
def main():
classes = [
'chimp',
'corvette',
'tokyo',
'goldengatebridge'
]
images, labels = get_labels(classes)
std_features = get_standard_features(images)
k = 256
surf_features = get_visual_words(images, k)
tas_features = get_tas_features(images)
feature_dict = {
'Std': std_features,
'SURF': surf_features,
'TAS': tas_features
#'Zernike': zernike_features
}
best_features = log_classify(feature_dict, labels)
classifier = LogisticRegression()
classifier.fit(best_features, labels)
def test_warm_start():
# A 1-iteration second fit on same data should give almost same result
# with warm starting, and quite different result without warm starting.
# Warm starting does not work with liblinear solver.
X, y = iris.data, iris.target
solvers = ['newton-cg', 'sag']
# old scipy doesn't have maxiter
if sp_version >= (0, 12):
solvers.append('lbfgs')
for warm_start in [True, False]:
for fit_intercept in [True, False]:
for solver in solvers:
for multi_class in ['ovr', 'multinomial']:
clf = LogisticRegression(tol=1e-4, multi_class=multi_class,
warm_start=warm_start,
solver=solver,
random_state=42, max_iter=100,
fit_intercept=fit_intercept)
clf.fit(X, y)
coef_1 = clf.coef_
clf.max_iter = 1
with ignore_warnings():
clf.fit(X, y)
cum_diff = np.sum(np.abs(coef_1 - clf.coef_))
msg = ("Warm starting issue with %s solver in %s mode "
"with fit_intercept=%s and warm_start=%s"
% (solver, multi_class, str(fit_intercept),
str(warm_start)))
if warm_start:
assert_greater(2.0, cum_diff, msg)
else:
assert_greater(cum_diff, 2.0, msg)
def test_write_parameters():
# Test that we can write to coef_ and intercept_
clf = LogisticRegression(random_state=0)
clf.fit(X, Y1)
clf.coef_[:] = 0
clf.intercept_[:] = 0
assert_array_almost_equal(clf.decision_function(X), 0)
def mlogistic():
X = []
# 前三行作为输入样本
X.append("f**k you")
X.append("f**k you all")
X.append("hello everyone")
# 后两句作为测试样本
X.append("f**k me")
X.append("hello boy")
# y为样本标注
y = [1,1,0]
vectorizer = TfidfVectorizer()
# 取X的前三句作为输入做tfidf转换
X_train = vectorizer.fit_transform(X[:-2])
print X_train
# 取X的后两句用“上句生成”的tfidf做转换
X_test = vectorizer.transform(X[-2:])
print X_test
# 用逻辑回归模型做训练
classifier = LogisticRegression()
classifier.fit(X_train, y)
# 做测试样例的预测
predictions = classifier.predict(X_test)
print predictions
def test_consistency_path():
"""Test that the path algorithm is consistent"""
rng = np.random.RandomState(0)
X = np.concatenate((rng.randn(100, 2) + [1, 1], rng.randn(100, 2)))
y = [1] * 100 + [-1] * 100
Cs = np.logspace(0, 4, 10)
f = ignore_warnings
# can't test with fit_intercept=True since LIBLINEAR
# penalizes the intercept
for method in ('lbfgs', 'newton-cg', 'liblinear'):
coefs, Cs = f(logistic_regression_path)(
X, y, Cs=Cs, fit_intercept=False, tol=1e-16, solver=method)
for i, C in enumerate(Cs):
lr = LogisticRegression(C=C, fit_intercept=False, tol=1e-16)
lr.fit(X, y)
lr_coef = lr.coef_.ravel()
assert_array_almost_equal(lr_coef, coefs[i], decimal=4)
# test for fit_intercept=True
for method in ('lbfgs', 'newton-cg', 'liblinear'):
Cs = [1e3]
coefs, Cs = f(logistic_regression_path)(
X, y, Cs=Cs, fit_intercept=True, tol=1e-4, solver=method)
lr = LogisticRegression(C=Cs[0], fit_intercept=True, tol=1e-4,
intercept_scaling=10000)
lr.fit(X, y)
lr_coef = np.concatenate([lr.coef_.ravel(), lr.intercept_])
assert_array_almost_equal(lr_coef, coefs[0], decimal=4)
def test_regularization_path(self):
# Check results using logistic path
num_samples = 10
num_feat = 5
X, y = make_classification(n_samples=num_samples, n_features=num_feat, n_informative=3,
n_classes=2, random_state=0, weights=[0.5, 0.5])
matrix = np.zeros((num_samples, num_feat + 2))
matrix[:,:-2] = X
matrix[:, -2] = np.ones(num_samples)
matrix[:, -1] = y
# Betas to test
logitfitL1 = LogisticRegressionL1()
lambda_grid = np.exp(-1 * np.linspace(1, 17, 200))
path = logitfitL1.fit(matrix, lambda_grid)
# Sklearn
cs = l1_min_c(X, y, loss='log') * np.logspace(0, 3)
# Computing regularization path using sklearn
clf = LogisticRegression(C=1.0, penalty='l1', tol=1e-6)
coefs_ = []
for c in cs:
clf.set_params(C=c)
clf.fit(X, y)
coefs_.append(clf.coef_.ravel().copy())
skbetas = np.append(clf.intercept_[0], clf.coef_)
np.testing.assert_almost_equal(skbetas, logitfitL1.coef_, 1)
def test_consistency_path():
# Test that the path algorithm is consistent
rng = np.random.RandomState(0)
X = np.concatenate((rng.randn(100, 2) + [1, 1], rng.randn(100, 2)))
y = [1] * 100 + [-1] * 100
Cs = np.logspace(0, 4, 10)
f = ignore_warnings
# can't test with fit_intercept=True since LIBLINEAR
# penalizes the intercept
for solver in ("lbfgs", "newton-cg", "liblinear", "sag"):
coefs, Cs, _ = f(logistic_regression_path)(
X, y, Cs=Cs, fit_intercept=False, tol=1e-5, solver=solver, random_state=0
)
for i, C in enumerate(Cs):
lr = LogisticRegression(C=C, fit_intercept=False, tol=1e-5, random_state=0)
lr.fit(X, y)
lr_coef = lr.coef_.ravel()
assert_array_almost_equal(lr_coef, coefs[i], decimal=4, err_msg="with solver = %s" % solver)
# test for fit_intercept=True
for solver in ("lbfgs", "newton-cg", "liblinear", "sag"):
Cs = [1e3]
coefs, Cs, _ = f(logistic_regression_path)(
X, y, Cs=Cs, fit_intercept=True, tol=1e-6, solver=solver, intercept_scaling=10000.0, random_state=0
)
lr = LogisticRegression(C=Cs[0], fit_intercept=True, tol=1e-4, intercept_scaling=10000.0, random_state=0)
lr.fit(X, y)
lr_coef = np.concatenate([lr.coef_.ravel(), lr.intercept_])
assert_array_almost_equal(lr_coef, coefs[0], decimal=4, err_msg="with solver = %s" % solver)
def predictWithThreshold(datadir, threshold, penalty_type='l2'):
maxent = LogisticRegression(penalty=penalty_type)
scores = defaultdict(list)
for dir in sorted(os.listdir(datadir), reverse=True):
trainfeatures, trainlabels, vec = feats_and_classify.collect_features(datadir+dir+'/train.conll')
TrainIndices=np.array(range(len(trainfeatures)))
features, labels, vec = feats_and_classify.collect_features(datadir+dir+'/all.conll')
TestIndices=np.array(range(len(trainfeatures),len(features)))
# print('\r'+dir, end="")
# print(dir)
TrainX_i = features[TrainIndices]
Trainy_i = labels[TrainIndices]
TestX_i = features[TestIndices]
Testy_i = labels[TestIndices]
maxent.fit(TrainX_i,Trainy_i)
# print('Finished fitting')
ypred_i, score=pred_for_threshold(maxent,TestX_i,Testy_i, threshold)
# print('Predicting')
scores["F1"].append(score[0])
scores["Recall"].append(score[1])
scores["Accuracy"].append(score[2])
scores["Precision"].append(score[3])
#scores = cross_validation.cross_val_score(maxent, features, labels, cv=10)
print("\n--")
for key in sorted(scores.keys()):
currentmetric = np.array(scores[key])
print("%s : %0.2f (+/- %0.2f)" % (key,currentmetric.mean(), currentmetric.std()))
print("--")
def main():
scriptdir = os.path.dirname(os.path.realpath(__file__))
parser = argparse.ArgumentParser(description="Skeleton for features and classifier for CWI-2016--optimisation of threshhold")
parser.add_argument('--threshold',type=float,default=0.5)
parser.add_argument('--annotator',type=str,default="03")
parser.add_argument('--penalty',type=str,choices=["l1","l2"],default="l1")
args = parser.parse_args()
current_single_ann = scriptdir+"/../data/cwi_training/cwi_training_"+args.annotator+".lbl.conll"
testfile = scriptdir+"/../data/cwi_testing/cwi_testing.txt.lbl.conll"
X__dict_train, y_train, v_train = feats_and_classify.collect_features(current_single_ann,vectorize=False)
X_dict_test, y_test, v_test = feats_and_classify.collect_features(testfile,vectorize=False)
featdicts = list([x for x in X__dict_train + X_dict_test])
vect = DictVectorizer()
X = vect.fit_transform(featdicts).toarray()
X_train=X[:len(y_train)]
X_test=X[len(y_train):]
maxent = LogisticRegression(penalty=args.penalty)
maxent.fit(X_train,y_train)
y_pred_proba = maxent.predict_proba(X_test)
ypred_i=["1" if pair[1]>=args.threshold else "0" for pair in y_pred_proba]
fout = open(args.annotator+".pred",mode="w")
print("\n".join(ypred_i),file=fout)
fout.close()
sys.exit(0)
def cvWithThreshold(X, y_current_tr, y_current_te, threshold, regularization='l2'):
out_dict = {}
scores = defaultdict(list)
fold=1
maxent = LogisticRegression(penalty=regularization)
for TrainIndices, TestIndices in cross_validation.StratifiedKFold(y_current_tr, n_folds=10, shuffle=False, random_state=None):
print('\r'+str(fold), end="")
fold+=1
TrainX_i = X[TrainIndices]
Trainy_i = y_current_tr[TrainIndices]
TestX_i = X[TestIndices]
Testy_i = y_current_te[TestIndices]
maxent.fit(TrainX_i,Trainy_i)
ypred_i, score=pred_for_threshold(maxent,TestX_i,Testy_i, threshold)
scores["F1"].append(score[0])
scores["Recall"].append(score[1])
scores["Accuracy"].append(score[2])
scores["Precision"].append(score[3])
#scores = cross_validation.cross_val_score(maxent, features, labels, cv=10)
print("\n--")
for key in sorted(scores.keys()):
currentmetric = np.array(scores[key])
out_dict[key] = (currentmetric.mean(),currentmetric.std())
print("%s : %0.2f (+/- %0.2f)" % (key,currentmetric.mean(), currentmetric.std()))
print("--")
return out_dict
def test_liblinear_random_state():
X, y = make_classification(n_samples=20)
lr1 = LogisticRegression(random_state=0)
lr1.fit(X, y)
lr2 = LogisticRegression(random_state=0)
lr2.fit(X, y)
assert_array_almost_equal(lr1.coef_, lr2.coef_)
def test_warm_start(solver, warm_start, fit_intercept, multi_class):
# A 1-iteration second fit on same data should give almost same result
# with warm starting, and quite different result without warm starting.
# Warm starting does not work with liblinear solver.
X, y = iris.data, iris.target
clf = LogisticRegression(tol=1e-4, multi_class=multi_class,
warm_start=warm_start,
solver=solver,
random_state=42, max_iter=100,
fit_intercept=fit_intercept)
with ignore_warnings(category=ConvergenceWarning):
clf.fit(X, y)
coef_1 = clf.coef_
clf.max_iter = 1
clf.fit(X, y)
cum_diff = np.sum(np.abs(coef_1 - clf.coef_))
msg = ("Warm starting issue with %s solver in %s mode "
"with fit_intercept=%s and warm_start=%s"
% (solver, multi_class, str(fit_intercept),
str(warm_start)))
if warm_start:
assert_greater(2.0, cum_diff, msg)
else:
assert_greater(cum_diff, 2.0, msg)
def test_logistic_regression_solvers():
X, y = make_classification(n_features=10, n_informative=5, random_state=0)
ncg = LogisticRegression(solver='newton-cg', fit_intercept=False)
lbf = LogisticRegression(solver='lbfgs', fit_intercept=False)
lib = LogisticRegression(fit_intercept=False)
sag = LogisticRegression(solver='sag', fit_intercept=False,
random_state=42)
saga = LogisticRegression(solver='saga', fit_intercept=False,
random_state=42)
ncg.fit(X, y)
lbf.fit(X, y)
sag.fit(X, y)
saga.fit(X, y)
lib.fit(X, y)
assert_array_almost_equal(ncg.coef_, lib.coef_, decimal=3)
assert_array_almost_equal(lib.coef_, lbf.coef_, decimal=3)
assert_array_almost_equal(ncg.coef_, lbf.coef_, decimal=3)
assert_array_almost_equal(sag.coef_, lib.coef_, decimal=3)
assert_array_almost_equal(sag.coef_, ncg.coef_, decimal=3)
assert_array_almost_equal(sag.coef_, lbf.coef_, decimal=3)
assert_array_almost_equal(saga.coef_, sag.coef_, decimal=3)
assert_array_almost_equal(saga.coef_, lbf.coef_, decimal=3)
assert_array_almost_equal(saga.coef_, ncg.coef_, decimal=3)
assert_array_almost_equal(saga.coef_, lib.coef_, decimal=3)
def test_logistic_regression_solvers_multiclass():
X, y = make_classification(n_samples=20, n_features=20, n_informative=10,
n_classes=3, random_state=0)
tol = 1e-7
ncg = LogisticRegression(solver='newton-cg', fit_intercept=False, tol=tol)
lbf = LogisticRegression(solver='lbfgs', fit_intercept=False, tol=tol)
lib = LogisticRegression(fit_intercept=False, tol=tol)
sag = LogisticRegression(solver='sag', fit_intercept=False, tol=tol,
max_iter=1000, random_state=42)
saga = LogisticRegression(solver='saga', fit_intercept=False, tol=tol,
max_iter=10000, random_state=42)
ncg.fit(X, y)
lbf.fit(X, y)
sag.fit(X, y)
saga.fit(X, y)
lib.fit(X, y)
assert_array_almost_equal(ncg.coef_, lib.coef_, decimal=4)
assert_array_almost_equal(lib.coef_, lbf.coef_, decimal=4)
assert_array_almost_equal(ncg.coef_, lbf.coef_, decimal=4)
assert_array_almost_equal(sag.coef_, lib.coef_, decimal=4)
assert_array_almost_equal(sag.coef_, ncg.coef_, decimal=4)
assert_array_almost_equal(sag.coef_, lbf.coef_, decimal=4)
assert_array_almost_equal(saga.coef_, sag.coef_, decimal=4)
assert_array_almost_equal(saga.coef_, lbf.coef_, decimal=4)
assert_array_almost_equal(saga.coef_, ncg.coef_, decimal=4)
assert_array_almost_equal(saga.coef_, lib.coef_, decimal=4)
def test_logreg_l1():
# Because liblinear penalizes the intercept and saga does not, we do not
# fit the intercept to make it possible to compare the coefficients of
# the two models at convergence.
rng = np.random.RandomState(42)
n_samples = 50
X, y = make_classification(n_samples=n_samples, n_features=20,
random_state=0)
X_noise = rng.normal(size=(n_samples, 3))
X_constant = np.ones(shape=(n_samples, 2))
X = np.concatenate((X, X_noise, X_constant), axis=1)
lr_liblinear = LogisticRegression(penalty="l1", C=1.0, solver='liblinear',
fit_intercept=False,
tol=1e-10)
lr_liblinear.fit(X, y)
lr_saga = LogisticRegression(penalty="l1", C=1.0, solver='saga',
fit_intercept=False,
max_iter=1000, tol=1e-10)
lr_saga.fit(X, y)
assert_array_almost_equal(lr_saga.coef_, lr_liblinear.coef_)
# Noise and constant features should be regularized to zero by the l1
# penalty
assert_array_almost_equal(lr_liblinear.coef_[0, -5:], np.zeros(5))
assert_array_almost_equal(lr_saga.coef_[0, -5:], np.zeros(5))
def test_logreg_cv_penalty():
# Test that the correct penalty is passed to the final fit.
X, y = make_classification(n_samples=50, n_features=20, random_state=0)
lr_cv = LogisticRegressionCV(penalty="l1", Cs=[1.0], solver='liblinear')
lr_cv.fit(X, y)
lr = LogisticRegression(penalty="l1", C=1.0, solver='liblinear')
lr.fit(X, y)
assert_equal(np.count_nonzero(lr_cv.coef_), np.count_nonzero(lr.coef_))
def classify_logistic(train_features, train_labels, test_features):
global SAVE
clf = LogisticRegression()
clf.fit(train_features, train_labels)
if not TEST and SAVE:
save_pickle("logistic", clf)
return clf.predict(test_features)
def main():
scriptdir = os.path.dirname(os.path.realpath(__file__))
default_pool = scriptdir+"/../data/cwi_training/cwi_training.txt.lbl.conll"
parser = argparse.ArgumentParser(description="Skeleton for features and classifier for CWI-2016--optimisation of threshhold")
parser.add_argument('--iterations',type=int,default=5)
args = parser.parse_args()
all_feats = []
all_labels = defaultdict(list)
scores = defaultdict(list)
for idx in "01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20".split(" "):
# for idx in "01".split(" "):
current_single_ann = scriptdir+"/../data/cwi_training/cwi_training_"+idx+".lbl.conll"
f_current, labels_current, v_current = feats_and_classify.collect_features(current_single_ann,vectorize=False,generateFeatures=False)
for instance_index,l in enumerate(labels_current):
all_labels[instance_index].append(l)
current_single_ann = scriptdir+"/../data/cwi_training/cwi_training_01.lbl.conll"
feats, labels_current, v_current = feats_and_classify.collect_features(current_single_ann,vectorize=True,generateFeatures=True)
for it in range(args.iterations):
for TrainIndices, TestIndices in cross_validation.KFold(n=feats.shape[0], n_folds=10, shuffle=True, random_state=None):
maxent = LogisticRegression(penalty='l2')
TrainX_i = feats[TrainIndices]
Trainy_i = [all_labels[x][random.randrange(0,20)] for x in TrainIndices]
TestX_i = feats[TestIndices]
Testy_i = [all_labels[x][random.randrange(0,20)] for x in TestIndices]
maxent.fit(TrainX_i,Trainy_i)
ypred_i = maxent.predict(TestX_i)
acc = accuracy_score(ypred_i, Testy_i)
pre = precision_score(ypred_i, Testy_i)
rec = recall_score(ypred_i, Testy_i)
# shared task uses f1 of *accuracy* and recall!
f1 = 2 * acc * rec / (acc + rec)
scores["Accuracy"].append(acc)
scores["F1"].append(f1)
scores["Precision"].append(pre)
scores["Recall"].append(rec)
#scores = cross_validation.cross_val_score(maxent, features, labels, cv=10)
print("--")
for key in sorted(scores.keys()):
currentmetric = np.array(scores[key])
print("%s : %0.2f (+/- %0.2f)" % (key,currentmetric.mean(), currentmetric.std()))
print("--")
sys.exit(0)
def my_module(rt, params, inputs, outputs):
# TODO : Fill your code here
X = pickle.load(open(inputs.X, 'r'))
Y = pickle.load(open(inputs.Y, 'r'))
model = LogisticRegression()
model.fit(X, Y)
pickle.dump(model, open(outputs.MODEL, 'w'))
print "Done"
def prepare(imageFolder):
print("preparing images...")
inputData = []
outputData = []
for file in os.listdir(imageFolder):
if (len(file) == 8):
inputData += getInputFromPic(Image.open(imageFolder + file).convert("L"), file)
outputData += getOutputFromFileName(file)
print("training model...")
model = LogisticRegression()
model.fit(inputData, outputData)
return model
def test_liblinear_decision_function_zero():
# Test negative prediction when decision_function values are zero.
# Liblinear predicts the positive class when decision_function values
# are zero. This is a test to verify that we do not do the same.
# See Issue: https://github.com/scikit-learn/scikit-learn/issues/3600
# and the PR https://github.com/scikit-learn/scikit-learn/pull/3623
X, y = make_classification(n_samples=5, n_features=5, random_state=0)
clf = LogisticRegression(fit_intercept=False)
clf.fit(X, y)
# Dummy data such that the decision function becomes zero.
X = np.zeros((5, 5))
assert_array_equal(clf.predict(X), np.zeros(5))
def fit_model_2(self, lol = .07, toWrite = False):
model = LogisticRegression(C = lol, penalty = 'l1', tol = 1e-6)
for data in self.cv_data:
X_train, X_test, Y_train, Y_test = data
X_train,Y_train = self.balance_data(X_train,Y_train)
model.fit(X_train,Y_train)
pred = model.predict_proba(X_test)[:,1]
print("Model 2 Score: %f" % (logloss(Y_test,pred),))
if toWrite:
f2 = open('model2/model.pkl','w')
pickle.dump(model,f2)
f2.close()
def test_logistic_regression_class_weights():
# Multinomial case: remove 90% of class 0
X = iris.data[45:, :]
y = iris.target[45:]
solvers = ("lbfgs", "newton-cg")
class_weight_dict = _compute_class_weight_dictionary(y)
for solver in solvers:
clf1 = LogisticRegression(solver=solver, multi_class="multinomial",
class_weight="balanced")
clf2 = LogisticRegression(solver=solver, multi_class="multinomial",
class_weight=class_weight_dict)
clf1.fit(X, y)
clf2.fit(X, y)
assert_array_almost_equal(clf1.coef_, clf2.coef_, decimal=4)
# Binary case: remove 90% of class 0 and 100% of class 2
X = iris.data[45:100, :]
y = iris.target[45:100]
solvers = ("lbfgs", "newton-cg", "liblinear")
class_weight_dict = _compute_class_weight_dictionary(y)
for solver in solvers:
clf1 = LogisticRegression(solver=solver, multi_class="ovr",
class_weight="balanced")
clf2 = LogisticRegression(solver=solver, multi_class="ovr",
class_weight=class_weight_dict)
clf1.fit(X, y)
clf2.fit(X, y)
assert_array_almost_equal(clf1.coef_, clf2.coef_, decimal=6)
def test_max_iter():
# Test that the maximum number of iteration is reached
X, y_bin = iris.data, iris.target.copy()
y_bin[y_bin == 2] = 0
solvers = ["newton-cg", "liblinear", "sag"]
# old scipy doesn't have maxiter
if sp_version >= (0, 12):
solvers.append("lbfgs")
for max_iter in range(1, 5):
for solver in solvers:
lr = LogisticRegression(max_iter=max_iter, tol=1e-15, random_state=0, solver=solver)
lr.fit(X, y_bin)
assert_equal(lr.n_iter_[0], max_iter)
def test_multinomial_binary_probabilities():
# Test multinomial LR gives expected probabilities based on the
# decision function, for a binary problem.
X, y = make_classification()
clf = LogisticRegression(multi_class='multinomial', solver='saga')
clf.fit(X, y)
decision = clf.decision_function(X)
proba = clf.predict_proba(X)
expected_proba_class_1 = (np.exp(decision) /
(np.exp(decision) + np.exp(-decision)))
expected_proba = np.c_[1-expected_proba_class_1, expected_proba_class_1]
assert_almost_equal(proba, expected_proba)
def clazzify(train_mat, test_mat, true_train_labels):
"""
"""
# learn
logging.info('learning...')
model = LogisticRegression(random_state=17, penalty='l1')
model.fit(train_mat, true_train_labels)
logging.info('finished learning.')
# test
logging.info('testing')
predicted_test_labels = model.predict(test_mat)
logging.info('finished testing')
return predicted_test_labels, model
def test_scikit_learn_exploded_data(self):
# Check results with scikit learn
betas = [0.001, 0.07, 0.4]
matrix = create_random_observations(200, 2, betas)
new_matrix = explode_matrix(matrix)
X = new_matrix[:,:-2]
y = new_matrix[:, -1]
lib = LogisticRegression(fit_intercept=True)
lib.fit(X, y)
path = self.logitfitL1.fit(new_matrix, self.lambda_grid)
skbetas = np.append(lib.intercept_[0], lib.coef_)
np.testing.assert_almost_equal(skbetas, self.logitfitL1.coef_, 2)
def generate_submission():
global alg, predictions, submission
# The columns we'll use to predict the target
# Initialize the algorithm class
alg = LogisticRegression(random_state=1)
# Train the algorithm using all the training data
alg.fit(train[predictors], train["Survived"])
# Make predictions using the test set.
predictions = alg.predict(test[predictors])
# Create a new dataframe with only the columns Kaggle wants from the dataset.
submission = pandas.DataFrame({
"PassengerId": test["PassengerId"],
"Survived": predictions
})
submission.to_csv("kaggle.csv", index=False)
print("kaggele.csv is generated")
def test_multinomial_binary():
"""Test multinomial LR on a binary problem."""
target = (iris.target > 0).astype(np.intp)
target = np.array(["setosa", "not-setosa"])[target]
clf = LogisticRegression(solver='lbfgs', multi_class='multinomial')
clf.fit(iris.data, target)
assert_equal(clf.coef_.shape, (1, iris.data.shape[1]))
assert_equal(clf.intercept_.shape, (1,))
assert_array_equal(clf.predict(iris.data), target)
mlr = LogisticRegression(solver='lbfgs', multi_class='multinomial',
fit_intercept=False)
mlr.fit(iris.data, target)
pred = clf.classes_[np.argmax(clf.predict_log_proba(iris.data), axis=1)]
assert_greater(np.mean(pred == target), .9)
def test_multinomial_logistic_regression_string_inputs():
# Test with string labels for LogisticRegression(CV)
n_samples, n_features, n_classes = 50, 5, 3
X_ref, y = make_classification(n_samples=n_samples,
n_features=n_features,
n_classes=n_classes,
n_informative=3)
y_str = LabelEncoder().fit(['bar', 'baz', 'foo']).inverse_transform(y)
# For numerical labels, let y values be taken from set (-1, 0, 1)
y = np.array(y) - 1
# Test for string labels
lr = LogisticRegression(solver='lbfgs', multi_class='multinomial')
lr_cv = LogisticRegressionCV(solver='lbfgs', multi_class='multinomial')
lr_str = LogisticRegression(solver='lbfgs', multi_class='multinomial')
lr_cv_str = LogisticRegressionCV(solver='lbfgs', multi_class='multinomial')
lr.fit(X_ref, y)
lr_cv.fit(X_ref, y)
lr_str.fit(X_ref, y_str)
lr_cv_str.fit(X_ref, y_str)
assert_array_almost_equal(lr.coef_, lr_str.coef_)
assert_equal(sorted(lr_str.classes_), ['bar', 'baz', 'foo'])
assert_array_almost_equal(lr_cv.coef_, lr_cv_str.coef_)
assert_equal(sorted(lr_str.classes_), ['bar', 'baz', 'foo'])
assert_equal(sorted(lr_cv_str.classes_), ['bar', 'baz', 'foo'])
# The predictions should be in original labels
assert_equal(sorted(np.unique(lr_str.predict(X_ref))),
['bar', 'baz', 'foo'])
assert_equal(sorted(np.unique(lr_cv_str.predict(X_ref))),
['bar', 'baz', 'foo'])
# Make sure class weights can be given with string labels
lr_cv_str = LogisticRegression(solver='lbfgs',
class_weight={
'bar': 1,
'baz': 2,
'foo': 0
},
multi_class='multinomial').fit(
X_ref, y_str)
assert_equal(sorted(np.unique(lr_cv_str.predict(X_ref))), ['bar', 'baz'])
def test_logreg_predict_proba_multinomial():
X, y = make_classification(n_samples=10, n_features=20, random_state=0,
n_classes=3, n_informative=10)
# Predicted probabilites using the true-entropy loss should give a
# smaller loss than those using the ovr method.
clf_multi = LogisticRegression(multi_class="multinomial", solver="lbfgs")
clf_multi.fit(X, y)
clf_multi_loss = log_loss(y, clf_multi.predict_proba(X))
clf_ovr = LogisticRegression(multi_class="ovr", solver="lbfgs")
clf_ovr.fit(X, y)
clf_ovr_loss = log_loss(y, clf_ovr.predict_proba(X))
assert_greater(clf_ovr_loss, clf_multi_loss)
# Predicted probabilites using the soft-max function should give a
# smaller loss than those using the logistic function.
clf_multi_loss = log_loss(y, clf_multi.predict_proba(X))
clf_wrong_loss = log_loss(y, clf_multi._predict_proba_lr(X))
assert_greater(clf_wrong_loss, clf_multi_loss)
def test_multinomial_binary():
# Test multinomial LR on a binary problem.
target = (iris.target > 0).astype(np.intp)
target = np.array(["setosa", "not-setosa"])[target]
for solver in ['lbfgs', 'newton-cg']:
clf = LogisticRegression(solver=solver, multi_class='multinomial')
clf.fit(iris.data, target)
assert_equal(clf.coef_.shape, (1, iris.data.shape[1]))
assert_equal(clf.intercept_.shape, (1,))
assert_array_equal(clf.predict(iris.data), target)
mlr = LogisticRegression(solver=solver, multi_class='multinomial',
fit_intercept=False)
mlr.fit(iris.data, target)
pred = clf.classes_[np.argmax(clf.predict_log_proba(iris.data),
axis=1)]
assert_greater(np.mean(pred == target), .9)
def train():
start = timeit.default_timer()
df = pd.read_csv('train.csv', delimiter=',', header=None)
print("start training ")
train, test = train_test_split(df, test_size=0.2)
X_train_raw = train[1]
y_toxic = train[2]
#y_severe_toxic = train[3]
#y_obscene = train[4]
#y_threat = train[5]
#y_insult = train[6]
#y_identity_hate = train[7]
vectorizer = TfidfVectorizer()
X_train = vectorizer.fit_transform(X_train_raw)
#classifier_toxic = LogisticRegression(solver ='lbfgs')
#classifier_toxic.fit(X_train, y_toxic)
# classifier_toxic = OneVsOneClassifier(LinearSVC(C=1.0, class_weight=None, dual=True, fit_intercept=True,
#intercept_scaling=1, loss='l2', multi_class='ovr', penalty='l2',
#random_state=None, tol=0.0001, verbose=0))
classifier_toxic = LogisticRegression(random_state=0, multi_class='ovr')
classifier_toxic.fit(X_train, y_toxic)
f = open("logistic_complain_vectorizer.pickle", 'wb')
pickle.dump(vectorizer, f)
f.close()
f = open("logistic_complain_classfier.pickle", 'wb')
pickle.dump(classifier_toxic, f)
f.close()
stop = timeit.default_timer()
print("training time took was : ")
print stop - start
def logistic_regression(inp):
arr = xml_parser.xml_parser('logistic_regression_Params.xml', inp)
n = len(arr)
if n != 14:
return False
try:
X, y = make_classification(n_samples=arr[0],
n_features=arr[1],
n_informative=arr[2],
n_classes=arr[3])
except ValueError:
# pass
# print("here")
return False
print(arr)
# print("here")
# try:
# with parallel_backend(backend):
# print("here0")
try:
clf = LogisticRegression(penalty=arr[6],
dual=arr[7],
tol=arr[8],
C=arr[9],
fit_intercept=arr[10],
intercept_scaling=arr[11],
solver=arr[5],
n_jobs=arr[4],
max_iter=arr[12],
multi_class=arr[13])
clf.fit(X, y)
# print("here1")
except ValueError:
# pass
# print("here2")
return False
# except KeyError:
# # print("here3")
# return False
return True
def test_max_iter():
# Test that the maximum number of iteration is reached
X, y_bin = iris.data, iris.target.copy()
y_bin[y_bin == 2] = 0
solvers = ['newton-cg', 'liblinear', 'sag']
# old scipy doesn't have maxiter
if sp_version >= (0, 12):
solvers.append('lbfgs')
for max_iter in range(1, 5):
for solver in solvers:
for multi_class in ['ovr', 'multinomial']:
if solver == 'liblinear' and multi_class == 'multinomial':
continue
lr = LogisticRegression(max_iter=max_iter, tol=1e-15,
multi_class=multi_class,
random_state=0, solver=solver)
lr.fit(X, y_bin)
assert_equal(lr.n_iter_[0], max_iter)
def main():
classes = ['chimp', 'corvette', 'tokyo', 'goldengatebridge']
images, labels = get_labels(classes)
std_features = get_standard_features(images)
k = 256
surf_features = get_visual_words(images, k)
tas_features = get_tas_features(images)
feature_dict = {
'Std': std_features,
'SURF': surf_features,
'TAS': tas_features
#'Zernike': zernike_features
}
best_features = log_classify(feature_dict, labels)
classifier = LogisticRegression()
classifier.fit(best_features, labels)
def getBestThreshold(features, labels_pooled, labels_current):
print("length of pooled and current", len(labels_pooled),
len(labels_current))
maxent = LogisticRegression(penalty='l1')
scores = {"F1": [], "Recall": [], "Accuracy": [], "Precision": []}
thresholds = []
print('Finding best thresholds...')
fold = 1
# for TrainIndices, TestIndices in cross_validation.StratifiedKFold(labels_pooled, n_folds=2, shuffle=False, random_state=None):
for TrainIndices, TestIndices in cross_validation.StratifiedKFold(
labels_pooled, n_folds=10, shuffle=False, random_state=None):
# for TrainIndices, TestIndices in cross_validation.KFold(n=features.shape[0], n_folds=10, shuffle=False, random_state=None):
print('\r' + str(fold), end="")
fold += 1
TrainX_i = features[TrainIndices]
Trainy_i = labels_pooled[TrainIndices]
TestX_i = features[TestIndices]
Testy_i = labels_current[TestIndices]
maxent.fit(TrainX_i, Trainy_i)
#get prediction
thresh_i, ypred_i, score = optimize_threshold(maxent, TestX_i, Testy_i)
thresholds.append(thresh_i)
scores["F1"].append(score[0])
scores["Recall"].append(score[1])
scores["Accuracy"].append(score[2])
scores["Precision"].append(score[3])
#scores = cross_validation.cross_val_score(maxent, features, labels, cv=10)
print("\n--")
for key in sorted(scores.keys()):
currentmetric = np.array(scores[key])
print("%s : %0.2f (+/- %0.2f)" %
(key, currentmetric.mean(), currentmetric.std()))
print("--")
return maxent, np.array(thresholds)
def getBestThreshold(X, y_current_tr, y_current_te, regularization='l2'):
assert len(X) == len(y_current_tr) == len(
y_current_te
), 'Number of features ({}), annotator1 labels ({}) and annotator2 labels ({}) is not equal!'.format(
len(X), len(y_current_tr), len(y_current_te))
maxent = LogisticRegression(penalty=regularization)
scores = {"F1": [], "Recall": [], "Accuracy": [], "Precision": []}
thresholds = []
print('Finding best thresholds...')
fold = 1
for TrainIndices, TestIndices in cross_validation.StratifiedKFold(
y_current_tr, n_folds=10, shuffle=False, random_state=None):
print('\r' + str(fold), end="")
fold += 1
TrainX_i = X[TrainIndices]
Trainy_i = y_current_tr[TrainIndices]
TestX_i = X[TestIndices]
Testy_i = y_current_te[TestIndices]
maxent.fit(TrainX_i, Trainy_i)
#get prediction
thresh_i, ypred_i, score = optimize_threshold(maxent, TestX_i, Testy_i)
thresholds.append(thresh_i)
scores["F1"].append(score[0])
scores["Recall"].append(score[1])
scores["Accuracy"].append(score[2])
scores["Precision"].append(score[3])
#scores = cross_validation.cross_val_score(maxent, features, labels, cv=10)
print("\n--")
for key in sorted(scores.keys()):
currentmetric = np.array(scores[key])
print("%s : %0.2f (+/- %0.2f)" %
(key, currentmetric.mean(), currentmetric.std()))
print("--")
return maxent, np.array(thresholds)
def predictions():
import numpy as np
import pandas as pd
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model.logistic import LogisticRegression
from sklearn.cross_validation import train_test_split, cross_val_score
df = pd.read_csv('data/SMSSpamCollection', delimiter='\t', header=None)
X_train_raw, X_test_raw, y_train, y_test = train_test_split(df[1], df[0])
#75% goes to training set and 25% goes to test set
vactorizer = TfidfVectorizer()
X_train = vactorizer.fit_transform(X_train_raw)
X_test = vactorizer.transform(X_test_raw)
#the model
classifier = LogisticRegression()
classifier.fit(X_train, y_train)
predictions = classifier.predict(X_test)
for i, prediction in enumerate(predictions[:5]):
print 'Prediction: %s. Message: %s' % (prediction, X_test_raw[i])
def regress_on_words(self, word_index, X):
"""
word: The word that we are interested in
text_corpus: input
"""
labels = []
tmp_X = X # Avoid directly changing the variable
for idx, sentence in enumerate(X):
if (sentence[word_index] == 1):
# tmp_X[idx][word_index] = 0
labels.append(1)
else:
labels.append(0)
# Build the logistic regression model
log_reg = LogisticRegression()
log_reg.fit(tmp_X, labels)
probs = log_reg.predict_proba(tmp_X)[:, -1]
return probs
def validate_model(X, y, N, digit, classifier):
"""
This function validate the model by K-fold cross validation and print the ROC curves
in the result folder
:param X: nparray, one row is one sample
:param y: nparray, labels
:param out: output filename
:param N: number of CPU cores to use
:param digit: digit of the captcha
:param classifier: which classifier to use
:return: None
"""
# K-fold cross validation
folds = KFold(n_splits=5, shuffle=True, random_state=1234567).split(X)
fold_r = fold_result()
labels = np.unique(y)
category_rs = [None] * len(labels)
for label in labels:
category_rs[label] = category_result()
for train, test in folds:
X_train = X[train]
X_test = X[test]
y_train = y[train]
y_test = y[test]
if (classifier == 'Logistic'):
clf = LogisticRegression(solver='sag', n_jobs=N)
else:
clf = RandomForestClassifier(n_estimators=200,
random_state=1234567,
n_jobs=N)
clf.fit(X_train, y_train)
probas = clf.predict_proba(X_test)
fold_r.append(clf, X_train, y_train, X_test, y_test)
for label in labels:
category_rs[label].append(y_test, probas[:, label], label)
fold_r.print_score(digit)
# print and save ROCS
for label in labels:
category_rs[label].print_result(label, digit)
def test_dtype_match():
# Test that np.float32 input data is not cast to np.float64 when possible
X_32 = np.array(X).astype(np.float32)
y_32 = np.array(Y1).astype(np.float32)
X_64 = np.array(X).astype(np.float64)
y_64 = np.array(Y1).astype(np.float64)
X_sparse_32 = sp.csr_matrix(X, dtype=np.float32)
for solver in ['newton-cg']:
for multi_class in ['ovr', 'multinomial']:
# Check type consistency
lr_32 = LogisticRegression(solver=solver, multi_class=multi_class)
lr_32.fit(X_32, y_32)
assert_equal(lr_32.coef_.dtype, X_32.dtype)
# check consistency with sparsity
lr_32_sparse = LogisticRegression(solver=solver,
multi_class=multi_class)
lr_32_sparse.fit(X_sparse_32, y_32)
assert_equal(lr_32_sparse.coef_.dtype, X_sparse_32.dtype)
# Check accuracy consistency
lr_64 = LogisticRegression(solver=solver, multi_class=multi_class)
lr_64.fit(X_64, y_64)
assert_equal(lr_64.coef_.dtype, X_64.dtype)
assert_almost_equal(lr_32.coef_, lr_64.coef_.astype(np.float32))
def weighted_params():
with open('../../../../baselines/GSBC/YELPNYC/finalgrps.json', 'r') as fp:
grps = json.load(fp)
X = []
Y = []
c = 0
tot = 0
mc = 0
# for grp in grps:
# scorepred=grps[grp]['scorepred']
# if len(grps[grp]['users'])>1 and (sum(scorepred[:5]))/6.0>0.4:
# tot=tot+1
for grp in grps:
scorepred = grps[grp]['scorepred']
if len(grps[grp]['users']) > 1:
mc = mc + 1
# mc=len(grps)
for grp in grps:
scorepred = grps[grp]['scorepred']
if len(grps[grp]['users']) > 1:
X.append(grps[grp]['scorepred'][:-2])
if c < int(0.72 * (mc)):
Y.append(0)
else:
Y.append(1)
c = c + 1
classifier = LogisticRegression()
classifier.fit(X, Y)
for grp in grps:
scorepred = grps[grp]['scorepred']
if len(grps[grp]['users']) > 1:
if grps[grp]['id'] not in ccgroups:
ccgroups[grps[grp]['id']] = 0
gtgroups[grps[grp]['id']] = 0
ccgroups[grps[grp]['id']] = sum([
grps[grp]['scorepred'][i] * classifier.coef_[0][i]
for i in range(8)
]) / 8.0
gtgroups[grps[grp]['id']] = grps[grp]['scoregt']
def test_logreg_l1_sparse_data():
# Because liblinear penalizes the intercept and saga does not, we do not
# fit the intercept to make it possible to compare the coefficients of
# the two models at convergence.
rng = np.random.RandomState(42)
n_samples = 50
X, y = make_classification(n_samples=n_samples, n_features=20,
random_state=0)
X_noise = rng.normal(scale=0.1, size=(n_samples, 3))
X_constant = np.zeros(shape=(n_samples, 2))
X = np.concatenate((X, X_noise, X_constant), axis=1)
X[X < 1] = 0
X = sparse.csr_matrix(X)
lr_liblinear = LogisticRegression(penalty="l1", C=1.0, solver='liblinear',
fit_intercept=False,
tol=1e-10)
lr_liblinear.fit(X, y)
lr_saga = LogisticRegression(penalty="l1", C=1.0, solver='saga',
fit_intercept=False,
max_iter=1000, tol=1e-10)
lr_saga.fit(X, y)
assert_array_almost_equal(lr_saga.coef_, lr_liblinear.coef_)
# Noise and constant features should be regularized to zero by the l1
# penalty
assert_array_almost_equal(lr_liblinear.coef_[0, -5:], np.zeros(5))
assert_array_almost_equal(lr_saga.coef_[0, -5:], np.zeros(5))
# Check that solving on the sparse and dense data yield the same results
lr_saga_dense = LogisticRegression(penalty="l1", C=1.0, solver='saga',
fit_intercept=False,
max_iter=1000, tol=1e-10)
lr_saga_dense.fit(X.toarray(), y)
assert_array_almost_equal(lr_saga.coef_, lr_saga_dense.coef_)
def gbdt_lr_train(libsvmFileName):
# load样本数据
X_all, y_all = load_svmlight_file(libsvmFileName)
# 训练/测试数据分割
X_train, X_test, y_train, y_test = train_test_split(X_all, y_all, test_size = 0.3, random_state = 42)
# 定义GBDT模型
gbdt = GradientBoostingClassifier(n_estimators=40, max_depth=3, verbose=0,max_features=0.5)
# 训练学习
gbdt.fit(X_train, y_train)
# 预测及AUC评测
y_pred_gbdt = gbdt.predict_proba(X_test.toarray())[:, 1]
gbdt_auc = roc_auc_score(y_test, y_pred_gbdt)
print('gbdt auc: %.5f' % gbdt_auc)
# lr对原始特征样本模型训练
lr = LogisticRegression()
lr.fit(X_train, y_train) # 预测及AUC评测
y_pred_test = lr.predict_proba(X_test)[:, 1]
lr_test_auc = roc_auc_score(y_test, y_pred_test)
print('基于原有特征的LR AUC: %.5f' % lr_test_auc)
# GBDT编码原有特征
X_train_leaves = gbdt.apply(X_train)[:,:,0]
X_test_leaves = gbdt.apply(X_test)[:,:,0]
# 对所有特征进行ont-hot编码
(train_rows, cols) = X_train_leaves.shape
gbdtenc = OneHotEncoder()
X_trans = gbdtenc.fit_transform(np.concatenate((X_train_leaves, X_test_leaves), axis=0))
# 定义LR模型
lr = LogisticRegression()
# lr对gbdt特征编码后的样本模型训练
lr.fit(X_trans[:train_rows, :], y_train)
# 预测及AUC评测
y_pred_gbdtlr1 = lr.predict_proba(X_trans[train_rows:, :])[:, 1]
gbdt_lr_auc1 = roc_auc_score(y_test, y_pred_gbdtlr1)
print('基于GBDT特征编码后的LR AUC: %.5f' % gbdt_lr_auc1)
# 定义LR模型
lr = LogisticRegression(n_jobs=-1)
# 组合特征
X_train_ext = hstack([X_trans[:train_rows, :], X_train])
X_test_ext = hstack([X_trans[train_rows:, :], X_test])
print(X_train_ext.shape)
# lr对组合特征的样本模型训练
lr.fit(X_train_ext, y_train)
# 预测及AUC评测
y_pred_gbdtlr2 = lr.predict_proba(X_test_ext)[:, 1]
gbdt_lr_auc2 = roc_auc_score(y_test, y_pred_gbdtlr2)
print('基于组合特征的LR AUC: %.5f' % gbdt_lr_auc2)
def _randomized_logistic(X, y, weights, mask, C=1., verbose=False,
fit_intercept=True, tol=1e-3):
X = X[safe_mask(X, mask)]
y = y[mask]
if issparse(X):
size = len(weights)
weight_dia = sparse.dia_matrix((1 - weights, 0), (size, size))
X = X * weight_dia
else:
X *= (1 - weights)
C = np.atleast_1d(np.asarray(C, dtype=np.float64))
scores = np.zeros((X.shape[1], len(C)), dtype=np.bool)
for this_C, this_scores in zip(C, scores.T):
# XXX : would be great to do it with a warm_start ...
clf = LogisticRegression(C=this_C, tol=tol, penalty='l1', dual=False,
fit_intercept=fit_intercept)
clf.fit(X, y)
this_scores[:] = np.any(
np.abs(clf.coef_) > 10 * np.finfo(np.float).eps, axis=0)
return scores
def main():
X = df_train.drop(['cust_id', 'y', 'cust_group'], axis=1, inplace=False)
y = df_train['y']
X_train,X_test , y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
print(X_train.shape, X_test.shape)
# X_train=extract_feature(X_train,y_train)
clf=LogisticRegression(C=1.0,max_iter=100,random_state=10)
print("===="*20)
clf.fit(X_train, y_train)
prob = clf.predict_proba(X_test)
pred = np.argmax(prob, axis=1)
print("mean_squared_error:", mean_squared_error(y_test, prob[:, 1]))
print("log_loss:", log_loss(y_test.astype(int), prob[:, 1]))
print("roc_auc_score：", roc_auc_score(y_test, prob[:, 1]))
# high_danger_prob=prob[:, 1]
# print(high_danger_prob)
# print("调参")
# tune_params(X_test, y_test)
predict(clf)
def test_consistency_path():
"""Test that the path algorithm is consistent"""
rng = np.random.RandomState(0)
X = np.concatenate((rng.randn(100, 2) + [1, 1], rng.randn(100, 2)))
y = [1] * 100 + [-1] * 100
Cs = np.logspace(0, 4, 10)
f = ignore_warnings
# can't test with fit_intercept=True since LIBLINEAR
# penalizes the intercept
for method in ('lbfgs', 'newton-cg', 'liblinear'):
coefs, Cs = f(logistic_regression_path)(X,
y,
Cs=Cs,
fit_intercept=False,
tol=1e-16,
solver=method)
for i, C in enumerate(Cs):
lr = LogisticRegression(C=C, fit_intercept=False, tol=1e-16)
lr.fit(X, y)
lr_coef = lr.coef_.ravel()
assert_array_almost_equal(lr_coef, coefs[i], decimal=4)
# test for fit_intercept=True
for method in ('lbfgs', 'newton-cg', 'liblinear'):
Cs = [1e3]
coefs, Cs = f(logistic_regression_path)(X,
y,
Cs=Cs,
fit_intercept=True,
tol=1e-4,
solver=method)
lr = LogisticRegression(C=Cs[0],
fit_intercept=True,
tol=1e-4,
intercept_scaling=10000)
lr.fit(X, y)
lr_coef = np.concatenate([lr.coef_.ravel(), lr.intercept_])
assert_array_almost_equal(lr_coef, coefs[0], decimal=4)
def logistic_model_by_sk_learn():
"""
使用sklearn实现
"""
data = pd.read_csv('./data/03_data_model/train_data_woe.csv')
# 应变量
Y = data['SeriousDlqin2yrs']
# 自变量，剔除对因变量影响不明显的变量
X = data.drop([
'SeriousDlqin2yrs', 'DebtRatio', 'MonthlyIncome',
'NumberOfOpenCreditLinesAndLoans', 'NumberRealEstateLoansOrLines',
'NumberOfDependents'
],
axis=1)
logistic = LogisticRegression()
logistic.fit(X,
Y,
sample_weight=(np.array(
X['RevolvingUtilizationOfUnsecuredLines'],
X['NumberOfTimes90DaysLate'])))
print(logistic.coef_[0])
return logistic
def _randomized_logistic(X,
y,
weights,
mask,
C=1.,
verbose=False,
fit_intercept=True,
tol=1e-3):
X = X[safe_mask(X, mask)]
y = y[mask]
if issparse(X):
size = len(weights)
weight_dia = sparse.dia_matrix((1 - weights, 0), (size, size))
X = X * weight_dia
else:
X *= (1 - weights)
C = np.atleast_1d(np.asarray(C, dtype=np.float64))
if C.ndim > 1:
raise ValueError(
"C should be 1-dimensional array-like, "
"but got a {}-dimensional array-like instead: {}.".format(
C.ndim, C))
scores = np.zeros((X.shape[1], len(C)), dtype=np.bool)
for this_C, this_scores in zip(C, scores.T):
# XXX : would be great to do it with a warm_start ...
clf = LogisticRegression(C=this_C,
tol=tol,
penalty='l1',
dual=False,
fit_intercept=fit_intercept,
solver='liblinear',
multi_class='ovr')
clf.fit(X, y)
this_scores[:] = np.any(
np.abs(clf.coef_) > 10 * np.finfo(np.float).eps, axis=0)
return scores
def test_warm_start():
# A 1-iteration second fit on same data should give almost same result
# with warm starting, and quite different result without warm starting.
# Warm starting does not work with liblinear solver.
X, y = iris.data, iris.target
solvers = ['newton-cg', 'sag']
# old scipy doesn't have maxiter
if sp_version >= (0, 12):
solvers.append('lbfgs')
for warm_start in [True, False]:
for fit_intercept in [True, False]:
for solver in solvers:
for multi_class in ['ovr', 'multinomial']:
if solver == 'sag' and multi_class == 'multinomial':
break
clf = LogisticRegression(tol=1e-4,
multi_class=multi_class,
warm_start=warm_start,
solver=solver,
random_state=42,
max_iter=100,
fit_intercept=fit_intercept)
clf.fit(X, y)
coef_1 = clf.coef_
clf.max_iter = 1
with ignore_warnings():
clf.fit(X, y)
cum_diff = np.sum(np.abs(coef_1 - clf.coef_))
msg = ("Warm starting issue with %s solver in %s mode "
"with fit_intercept=%s and warm_start=%s" %
(solver, multi_class, str(fit_intercept),
str(warm_start)))
if warm_start:
assert_greater(2.0, cum_diff, msg)
else:
assert_greater(cum_diff, 2.0, msg)
def test_saga_vs_liblinear():
iris = load_iris()
X, y = iris.data, iris.target
X = np.concatenate([X] * 10)
y = np.concatenate([y] * 10)
X_bin = X[y <= 1]
y_bin = y[y <= 1] * 2 - 1
X_sparse, y_sparse = make_classification(n_samples=50, n_features=20,
random_state=0)
X_sparse = sparse.csr_matrix(X_sparse)
for (X, y) in ((X_bin, y_bin), (X_sparse, y_sparse)):
for penalty in ['l1', 'l2']:
n_samples = X.shape[0]
# alpha=1e-3 is time consuming
for alpha in np.logspace(-1, 1, 3):
saga = LogisticRegression(
C=1. / (n_samples * alpha),
solver='saga',
multi_class='ovr',
max_iter=200,
fit_intercept=False,
penalty=penalty, random_state=0, tol=1e-24)
liblinear = LogisticRegression(
C=1. / (n_samples * alpha),
solver='liblinear',
multi_class='ovr',
max_iter=200,
fit_intercept=False,
penalty=penalty, random_state=0, tol=1e-24)
saga.fit(X, y)
liblinear.fit(X, y)
# Convergence for alpha=1e-3 is very slow
assert_array_almost_equal(saga.coef_, liblinear.coef_, 3)
class DetectMalicious():
def __init__(self):
benign_data = np.loadtxt('javascript-collection/benignjs.csv',
delimiter=',',
dtype=np.int32)
evil_data = np.loadtxt('javascript-collection/eviljs.csv',
delimiter=',',
dtype=np.int32)
train_data = np.concatenate((benign_data, evil_data), axis=0)
self.score_template = 'TPR %(TPR)f\tFPR %(FPR)f\tAccuracy %(Accuracy)f\tAUC %(AUC)f'
self.D = LogisticRegression()
self.D.fit(train_data[:, :-1], train_data[:, -1])
self.jsapi = []
for line in open('javascript-collection/jsapi.txt'):
self.jsapi.append(line.strip('\n'))
def predict(self, X):
flag = [0 for x in range(len(self.jsapi))]
for i in range(len(self.jsapi)):
if X.find(self.jsapi[i]) != -1:
flag[i] = 1
print self.D.predict([flag])
return self.D.predict([flag])
def getLRAcc(my_data, train_ratio):
total_rows = numpy.size(my_data, 0)
total_cols = numpy.size(my_data, 1)
train_rows = int(total_rows * train_ratio)
numpy.random.shuffle(my_data)
training, test = my_data[:train_rows, :], my_data[train_rows:, :]
XTrain = training[:, 1:total_cols - 1]
YTrain = training[:, total_cols - 1]
XTest = test[:, 1:total_cols - 1]
YTest = test[:, total_cols - 1]
lrMulti = LogisticRegression()
lrMulti.fit(XTrain, YTrain)
pred_Y = lrMulti.predict(XTest)
error = numpy.sum(YTest != pred_Y)
nb_accuracy = (float(total_rows - error) / total_rows) * 100
return nb_accuracy
def test_database_reconstruction_logistic_regression(get_iris_dataset):
(x_train_iris, y_train_iris), (x_test_iris, y_test_iris) = get_iris_dataset
y_train_iris = np.array([np.argmax(y) for y in y_train_iris])
y_test_iris = np.array([np.argmax(y) for y in y_test_iris])
x_private = x_test_iris[0, :].reshape(1, -1)
y_private = y_test_iris[0]
x_input = np.vstack((x_train_iris, x_private))
y_input = np.hstack((y_train_iris, y_private))
nb_private = LogisticRegression()
nb_private.fit(x_input, y_input)
estimator_private = ScikitlearnLogisticRegression(model=nb_private)
recon = DatabaseReconstruction(estimator=estimator_private)
x_recon, y_recon = recon.reconstruct(x_train_iris, y_train_iris)
assert x_recon is not None
assert x_recon.shape == (1, 4)
assert y_recon.shape == (1, 3)
assert np.isclose(x_recon, x_private, rtol=0.05).all()
assert np.argmax(y_recon, axis=1) == y_private
def test_logistic_regression_multinomial():
# Tests for the multinomial option in logistic regression
# Some basic attributes of Logistic Regression
n_samples, n_features, n_classes = 50, 20, 3
X, y = make_classification(n_samples=n_samples,
n_features=n_features,
n_informative=10,
n_classes=n_classes,
random_state=0)
clf_int = LogisticRegression(solver='lbfgs', multi_class='multinomial')
clf_int.fit(X, y)
assert_array_equal(clf_int.coef_.shape, (n_classes, n_features))
clf_wint = LogisticRegression(solver='lbfgs',
multi_class='multinomial',
fit_intercept=False)
clf_wint.fit(X, y)
assert_array_equal(clf_wint.coef_.shape, (n_classes, n_features))
# Similar tests for newton-cg solver option
clf_ncg_int = LogisticRegression(solver='newton-cg',
multi_class='multinomial')
clf_ncg_int.fit(X, y)
assert_array_equal(clf_ncg_int.coef_.shape, (n_classes, n_features))
clf_ncg_wint = LogisticRegression(solver='newton-cg',
fit_intercept=False,
multi_class='multinomial')
clf_ncg_wint.fit(X, y)
assert_array_equal(clf_ncg_wint.coef_.shape, (n_classes, n_features))
# Compare solutions between lbfgs and newton-cg
assert_almost_equal(clf_int.coef_, clf_ncg_int.coef_, decimal=3)
assert_almost_equal(clf_wint.coef_, clf_ncg_wint.coef_, decimal=3)
assert_almost_equal(clf_int.intercept_, clf_ncg_int.intercept_, decimal=3)
# Test that the path give almost the same results. However since in this
# case we take the average of the coefs after fitting across all the
# folds, it need not be exactly the same.
for solver in ['lbfgs', 'newton-cg']:
clf_path = LogisticRegressionCV(solver=solver,
multi_class='multinomial',
Cs=[1.])
clf_path.fit(X, y)
assert_array_almost_equal(clf_path.coef_, clf_int.coef_, decimal=3)
assert_almost_equal(clf_path.intercept_, clf_int.intercept_, decimal=3)
def cvWithThreshold(X,
y_current_tr,
y_current_te,
threshold,
regularization='l2'):
out_dict = {}
scores = defaultdict(list)
fold = 1
maxent = LogisticRegression(penalty=regularization)
for TrainIndices, TestIndices in cross_validation.StratifiedKFold(
y_current_tr, n_folds=10, shuffle=False, random_state=None):
print('\r' + str(fold), end="")
fold += 1
TrainX_i = X[TrainIndices]
Trainy_i = y_current_tr[TrainIndices]
TestX_i = X[TestIndices]
Testy_i = y_current_te[TestIndices]
maxent.fit(TrainX_i, Trainy_i)
ypred_i, score = pred_for_threshold(maxent, TestX_i, Testy_i,
threshold)
scores["F1"].append(score[0])
scores["Recall"].append(score[1])
scores["Accuracy"].append(score[2])
scores["Precision"].append(score[3])
#scores = cross_validation.cross_val_score(maxent, features, labels, cv=10)
print("\n--")
for key in sorted(scores.keys()):
currentmetric = np.array(scores[key])
out_dict[key] = (currentmetric.mean(), currentmetric.std())
print("%s : %0.2f (+/- %0.2f)" %
(key, currentmetric.mean(), currentmetric.std()))
print("--")
return out_dict
def predictWithThreshold(datadir, threshold, penalty_type='l2'):
maxent = LogisticRegression(penalty=penalty_type)
scores = defaultdict(list)
for dir in sorted(os.listdir(datadir), reverse=True):
trainfeatures, trainlabels, vec = feats_and_classify.collect_features(
datadir + dir + '/train.conll')
TrainIndices = np.array(range(len(trainfeatures)))
features, labels, vec = feats_and_classify.collect_features(
datadir + dir + '/all.conll')
TestIndices = np.array(range(len(trainfeatures), len(features)))
# print('\r'+dir, end="")
# print(dir)
TrainX_i = features[TrainIndices]
Trainy_i = labels[TrainIndices]
TestX_i = features[TestIndices]
Testy_i = labels[TestIndices]
maxent.fit(TrainX_i, Trainy_i)
# print('Finished fitting')
ypred_i, score = pred_for_threshold(maxent, TestX_i, Testy_i,
threshold)
# print('Predicting')
scores["F1"].append(score[0])
scores["Recall"].append(score[1])
scores["Accuracy"].append(score[2])
scores["Precision"].append(score[3])
#scores = cross_validation.cross_val_score(maxent, features, labels, cv=10)
print("\n--")
for key in sorted(scores.keys()):
currentmetric = np.array(scores[key])
print("%s : %0.2f (+/- %0.2f)" %
(key, currentmetric.mean(), currentmetric.std()))
print("--")
def gbdt_lr_train(self, Train_tab, Train_libsvm):
# load样本数据
X_all, y_all = load_svmlight_file("sample_libsvm_data.txt")
# 训练/测试数据分割
X_train, X_test, y_train, y_test = train_test_split(X_all,
y_all,
test_size=0.1,
random_state=42)
# 定义GBDT模型
gbdt = GradientBoostingClassifier(n_estimators=40,
max_depth=3,
verbose=0,
max_features=0.5)
# 训练模型
gbdt.fit(X_train, y_train)
# GBDT编码原有特征
X_train_leaves = gbdt.apply(X_train)[:, :, 0]
X_test_leaves = gbdt.apply(X_test)[:, :, 0]
# 对所有特征进行ont-hot编码
(train_rows, cols) = X_train_leaves.shape
gbdtenc = OneHotEncoder()
X_trans = gbdtenc.fit_transform(
np.concatenate((X_train_leaves, X_test_leaves), axis=0))
# 定义LR模型
lr = LogisticRegression(n_jobs=-1)
# 组合特征
X_train_ext = hstack([X_trans[:train_rows, :], X_train])
X_test_ext = hstack([X_trans[train_rows:, :], X_test])
# lr对组合特征的样本模型训练
lr.fit(X_train_ext, y_train)
# 预测及AUC评测
filename = 'finalized_model.sav'
pickle.dump(lr, open(filename, 'wb'))
# load the model from disk
loaded_model = pickle.load(open(filename, 'rb'))
y_pred_gbdtlr2 = loaded_model.predict_proba(X_test_ext)[:, 1]
print(y_pred_gbdtlr2)
