def test_multiclass():
dataset = datasets.load_wine()
oversampler = sv.MulticlassOversampling(sv.distance_SMOTE())
X_samp, y_samp = oversampler.sample(dataset['data'], dataset['target'])
assert len(X_samp) > 0
oversampler = sv.MulticlassOversampling(sv.distance_SMOTE(),
strategy='equalize_1_vs_many')
X_samp, y_samp = oversampler.sample(dataset['data'], dataset['target'])
def balanceData(self):
shape = self.trainInputDict["data"].shape
print("trainInputDict[data].shape : ", shape)
copy = self.trainInputDict["data"]
copy = copy.reshape(shape[0], -1)
print("copy.shape : ", copy.shape)
npDict = copy.numpy()
copyLabel = self.trainInputDict["label"]
print("copyLabel.shape : ", copyLabel.shape)
# copyLabel = copyLabel.view(-1)
npLabel = copyLabel.numpy()
# [print('Class {} had {} instances originally'.format(label, count)) for label, count in zip(*np.unique(npLabel, return_counts=True))]
# X_resampled, y_resampled = kmeans_smote.fit_sample(npDict, npLabel)
# print(sv.get_all_oversamplers_multiclass())
oversampler = sv.MulticlassOversampling(sv.SMOTE(n_jobs=6))
# oversampler = sv.SMOTE(n_jobs=8)
X_resampled, y_resampled = oversampler.sample(npDict, npLabel)
[
print('Class {} has {} instances after oversampling'.format(
label, count)) for label, count in zip(
*np.unique(y_resampled, return_counts=True))
]
newData = torch.from_numpy(
X_resampled.reshape(len(X_resampled), shape[1], shape[2],
shape[3]))
newLabel = torch.from_numpy(y_resampled)
newData = newData.float()
return newData, newLabel
def smote_data(images, labels):
# images = images[:50]
# labels = labels[:50]
shape = np.shape(images)
nums = shape[0] // 2
oversampler = sv.MulticlassOversampling(
sv.Borderline_SMOTE2(proportion=0.7,
n_neighbors=3,
k_neighbors=3,
n_jobs=12)) # MDO
X, y = oversampler.sample(np.reshape(images, (len(images), -1)), labels)
X = X.reshape((len(y), shape[1], shape[2], shape[3])).astype(np.uint8)
mkdir('new_train')
with open('new_train_label.csv', 'a', encoding='utf-8') as f:
f_csv = csv.writer(f)
for i, x in enumerate(X):
im = Image.fromarray(x)
im.save('./new_train/' + str(i) + '.jpg', 'jpeg')
f_csv.writerow([str(i) + '.jpg', y[i] + 1])
print('org: %d -> x: %d' % (len(labels), len(y)))
ys = [0] * 10
for i in y:
ys[i + 1] += 1
print(ys)
def over_sample(X, y, len_data, random_seed=None): # -> 没有使用
"""
对样本较少的类别进行过采样，增加样本数目，实现样本平衡
"""
oversampler = sv.MulticlassOversampling(
sv.distance_SMOTE(random_state=random_seed))
X_samp, y_samp = oversampler.sample(X, y)
return X_samp, y_samp
def generate_multiclass_figures():
oversamplers = sv.get_all_oversamplers()
oversamplers = [
o for o in oversamplers
if not sv.OverSampling.cat_changes_majority in o.categories
and 'proportion' in o().get_params()
]
import sklearn.datasets as datasets
dataset = datasets.load_wine()
X = dataset['data']
y = dataset['target']
import matplotlib.pyplot as plt
import sklearn.preprocessing as preprocessing
ss = preprocessing.StandardScaler()
X_ss = ss.fit_transform(X)
def plot_and_save(X, y, filename, oversampler_name):
plt.figure(figsize=(4, 3))
plt.scatter(X[y == 0][:, 0],
X[y == 0][:, 1],
c='r',
marker='o',
label='class 0')
plt.scatter(X[y == 1][:, 0],
X[y == 1][:, 1],
c='b',
marker='P',
label='class 1')
plt.scatter(X[y == 2][:, 0],
X[y == 2][:, 1],
c='green',
marker='x',
label='class 2')
plt.xlabel('feature 0')
plt.ylabel('feature 1')
plt.title(", ".join(["wine dataset", oversampler_name]))
plt.savefig(filename)
plt.show()
plot_and_save(X, y, 'figures/multiclass-base.png', "No Oversampling")
for o in oversamplers:
print(o.__name__)
mcos = sv.MulticlassOversampling(o())
X_samp, y_samp = mcos.sample(X_ss, y)
plot_and_save(ss.inverse_transform(X_samp), y_samp,
"figures/multiclass-%s" % o.__name__, o.__name__)
def oversampling_data(self):
for i in range(1, self.no_of_splits + 1):
train_data = np.load(
os.sep.join([self.saving_dir,
'Fold_{}_X_train.npy'.format(i)]))
train_data = train_data.reshape(-1, self.data_shape)
train_onehot = np.load(
os.sep.join([self.saving_dir,
'Fold_{}_y_train.npy'.format(i)]))
le_encoder = LabelEncoder()
y_train = le_encoder.fit_transform(train_onehot)
analysis_groups = self.groups_to_analyse
for analysis in analysis_groups:
group = self.Groups[analysis]
X = []
Y = []
try:
os.makedirs(
os.sep.join([
self.oversampled_data_dir, 'fold_' + str(i),
analysis
]))
except Exception as e:
print('Directory already exists')
for method in group.keys():
oversampler = sv.MulticlassOversampling(group[method])
x, y = oversampler.sample(train_data, y_train)
X.extend(x)
Y.extend(y)
x = pd.DataFrame(x)
x['label'] = y
x.to_csv(os.sep.join([
self.oversampled_data_dir, 'fold_' + str(i), analysis,
str(method) + '.csv'
]),
index=False)
X = pd.DataFrame(X)
X['label'] = Y
X.to_csv(os.sep.join([
self.oversampled_data_dir, 'fold_' + str(i), analysis,
str(analysis) + '.csv'
]),
index=False)
# In[4]:
# printing the number of samples
for i in np.unique(y):
print("class %d - samples: %d" % (i, np.sum(y == i)))
# ## Oversampling
#
# In this section multiclass oversampling is driven by the binary oversampler ```distance_SMOTE```.
# In[5]:
# chosing an oversampler supporting multiclass oversampling
oversampler = sv.MulticlassOversampling(sv.distance_SMOTE())
# In[6]:
X_samp, y_samp = oversampler.sample(X, y)
# ## Illustrating the outcome
# In[7]:
# printing the number of samples
for i in np.unique(y_samp):
print("class %d - samples: %d" % (i, np.sum(y_samp == i)))
# In[8]:
data_x_tst = pd.read_csv('nepal_earthquake_tst.csv')
# Comprueba balanceo de clases
# GraficoComprobarVar(data_y, "damage_grade")
# Comprueba valores perdidos
#ComprobarValPer(data_x)
# Calcula la matriz de correlación
#MatrizCorrelacion(data_x)
# Elimina etiquetas
eliminaLabels(data_x, data_y, data_x_tst, ['building_id'])
# Preprocesado category to number
X = catToNum(data_x).values
y = np.ravel(data_y.values)
X_tst = catToNum(data_x_tst).values
oversampler = sv.MulticlassOversampling(sv.distance_SMOTE(proportion=0.75))
X_sample, y_sample = oversampler.sample(X, y)
'''
print("------ RandomForest...")
rfm = RandomForestClassifier(max_features = 'sqrt', criterion='gini', n_estimators=500, \
max_depth=25, random_state=76592621, \
n_jobs=-1)
# Hago la validación cruzada para el algoritmo
#rfm, y_train_clf, y_test_clf = validacion_cruzada(rfm, X_sample, y_sample)
print("------ Generando submission...")
submission(X_sample, y_sample, X_tst, rfm)
'''
print("------ Catboost...")
cbc = CatBoostClassifier(n_estimators=450,
validator= RepeatedStratifiedKFold(n_repeats= 8,
n_splits= 5))
print(results.T[['sampler', 'auc', 'gacc']])
np.random.seed(random_seed)
results= sv.cross_validate(dataset= libras,
sampler= sv.NoSMOTE(),
classifier= KNeighborsClassifier(),
validator= RepeatedStratifiedKFold(n_repeats= 8,
n_splits= 5))
print(results.T[['sampler', 'auc', 'gacc']])
#%% running multiclass oversampling
np.random.seed(random_seed)
mc_oversampler= sv.MulticlassOversampling(sv.distance_SMOTE(), strategy= 'equalize_1_vs_many_successive')
X_os, y_os= mc_oversampler.sample(wine['data'], wine['target'])
plot_mc(wine['data'], wine['target'], 'wine', 0, 1, 2, 'wine.eps')
plot_mc(X_os, y_os, 'wine oversampled by distance-SMOTE', 0, 1, 2, 'wine_distance_smote.eps')
#%% oversampler evaluation
import os.path
ecoli['name']= 'ecoli'
cache_path= os.path.join(os.path.expanduser('~'), 'smote_cache')
np.random.seed(random_seed)
results= sv.evaluate_oversamplers(datasets= [ecoli],
samplers= [sv.SPY,
X, X_tst, y = PreprocesadoDatos(3, data_training, data_test, data_labels)
#Sustituir el valor de n por el modelo que se desea ejecutar
n=1
print("##################################")
print("Ejecutando algoritmo número " + str(n))
print("##################################")
import smote_variants as sv
#oversampler = sv.ProWSyn(proportion=1.0, n_neighbors=5,L=5, theta=1.0, n_jobs=-1, random_state=2)
#oversampler= sv.MulticlassOversampling(sv.distance_SMOTE(random_state=2))
oversampler = sv.MulticlassOversampling(sv.polynom_fit_SMOTE(topology='star',random_state=2))
# X_sam and y_sam contain the oversampled dataset
X_sam, y_sam= oversampler.sample(X, y)
'''
import collections
height=[collections.Counter(y_sam)[1],collections.Counter(y_sam)[2],collections.Counter(y_sam)[3]]
print(height)
print(collections.Counter(y))
plt.bar( ['low damage','medium damage','high damage'], height,color=['blue', 'orange', 'green'])
colors = {'class1 - Smote Polynom/star':'blue', 'class2 - Smote Polynom/star':'orange', 'class3 - Smote Polynom/star':'green'}
labels = list(colors.keys())
handles = [plt.Rectangle((1,0),-1,-1, color=colors[label]) for label in labels]