def train(df, fit_file):
print("Training...")
train_size = 0.75
vectorizer = CountVectorizer(analyzer="word",
tokenizer=None,
preprocessor=None,
stop_words=None)
clf = automl.AutoSklearnClassifier(include_preprocessors=[
"no_preprocessing",
],
exclude_preprocessors=None)
encoder = LabelEncoder()
y = df.categoria
y = encoder.fit_transform(y)
X = vectorizer.fit_transform(df.nome_desc)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
train_size=train_size)
clf.fit(X_train, y_train)
accuracy = clf.score(X_test, y_test)
msg = "Accuracy with {:.0%} of testing data: {:.1%}".format(
1 - train_size, accuracy)
print(msg)
joblib.dump(clf, fit_file)
joblib.dump(encoder, 'encoder_%s' % fit_file)
joblib.dump(vectorizer, 'vectorizer_%s' % fit_file)
def train_autosklearn(alldata, labels, mtype, jsonfile, problemtype, default_features):
foldername=jsonfile[0:-5]+'_autosklearn_%s'%(default_features)
X_train, X_test, y_train, y_test = train_test_split(alldata,
labels,
train_size=0.750,
test_size=0.250,
random_state=42,
shuffle=True)
feature_types = (['numerical'] * len(X_train[0]))
automl = asklc.AutoSklearnClassifier(
time_left_for_this_task=60,
per_run_time_limit=300,
ml_memory_limit=10240,
tmp_folder=os.getcwd()+'/'+foldername+'_tmp',
output_folder=os.getcwd()+'/'+foldername,
delete_tmp_folder_after_terminate=False,
delete_output_folder_after_terminate=False)
automl.fit(X_train,
y_train,
dataset_name=jsonfile[0:-5],
feat_type=feature_types)
y_predictions = automl.predict(X_test)
acc= sklearn.metrics.accuracy_score(y_true=y_test,
y_pred=y_predictions)
print("Accuracy:", acc)
print('saving classifier to disk')
f=open(modelname+'.pickle','wb')
pickle.dump(automl,f)
f.close()
data={'sample type': problemtype,
'training script': 'autosklearn',
'feature_set':default_features,
'model name':modelname+'.pickle',
'accuracy':acc,
'model type':'sc_'+classifiername,
}
g2=open(modelname+'.json','w')
json.dump(data,g2)
g2.close()
cur_dir2=os.getcwd()
try:
os.chdir(problemtype+'_models')
except:
os.mkdir(problemtype+'_models')
os.chdir(problemtype+'_models')
# now move all the files over to proper model directory
shutil.move(cur_dir2+'/'+modelname+'.json', os.getcwd()+'/'+modelname+'.json')
shutil.move(cur_dir2+'/'+modelname+'.pickle', os.getcwd()+'/'+modelname+'.pickle')
def execute(self, params, **kwargs):
import autosklearn.classification as automl
clf = automl.AutoSklearnClassifier(include_preprocessors=[
"no_preprocessing",
],
exclude_preprocessors=None)
clf.fit(self.marvin_dataset["X_train"], self.marvin_dataset["y_train"])
self.marvin_model = {
"clf": clf,
"vect": self.marvin_dataset["vect"],
"encoder": self.marvin_dataset["encoder"]
}
#
if alg.name == 'TPOT_Classifier':
from tpot import TPOTClassifier
model = TPOTClassifier(
generations=alg.generations,
cv=alg.cv,
scoring=alg.scoring,
verbosity=alg.verbosity
)
warn_not_gpu_support(alg)
elif alg.name == 'AutoSklearn_Classifier':
from autosklearn import classification
if alg.sampling:
model = classification.AutoSklearnClassifier(
time_left_for_this_task=alg.task_time,
per_run_time_limit=alg.run_time,
resampling_strategy=alg.sampling_strategy,
resampling_strategy_arguments={'folds': alg.folds}
)
else:
model = classification.AutoSklearnClassifier(
time_left_for_this_task=alg.task_time,
per_run_time_limit=alg.run_time
)
warn_not_gpu_support(alg)
elif alg.name == 'SupportVectorMachines':
if NVIDIA_RAPIDS_ENABLED:
from cuml.svm import SVC
else:
from sklearn.svm import SVC
model = SVC(**alg.input_variables.__dict__)
elif alg.name == 'GaussianNaiveBayes':
# Classification algorithms
#
if alg.name == 'TPOT_Classifier':
from tpot import TPOTClassifier
model = TPOTClassifier(
generations=alg.generations,
cv=alg.cv,
scoring=alg.scoring,
verbosity=alg.verbosity)
elif alg.name == 'AutoSklearn_Classifier':
from autosklearn import classification
if alg.sampling.lower()=='true':
model = classification.AutoSklearnClassifier(
time_left_for_this_task=alg.task_time,
per_run_time_limit=alg.run_time,
resampling_strategy= "".join(alg.sampling_strategy),
resampling_strategy_arguments={'folds':int(alg.folds)}
#feat_type = {Numerical,Numerical,Numerical,Numerical,Categorical}
)
else:
model = classification.AutoSklearnClassifier(
time_left_for_this_task=alg.task_time,
per_run_time_limit=alg.run_time
)
elif alg.name == 'SupportVectorMachines':
from sklearn.svm import SVC
model = SVC(**vars)
elif alg.name == 'GaussianNaiveBayes':
from sklearn.naive_bayes import GaussianNB
model = GaussianNB(**vars)
elif alg.name == 'LogisticRegression':
def main():
time = 60 * 5
runs = 5
option = int(input("1. Imbalanced\n2. SMOTE + ENN\n"))
for j in range(0, runs):
file = pd.read_csv("WA_Fn-UseC_-Telco-Customer-Churn.csv")
file = ut.prune_data(file)
X_train, X_test, Y_train, Y_test = sk.model_selection.train_test_split(
file.drop('Churn', axis=1), file['Churn'], test_size=0.3)
# Choosing kappa as the metric
kappas = asc_t.metrics.make_scorer('kappa',
sk.metrics.cohen_kappa_score)
if option == 1:
automl = asc.AutoSklearnClassifier(
time_left_for_this_task=time + 10,
per_run_time_limit=time,
initial_configurations_via_metalearning=0,
# resampling_strategy='cv',
# resampling_strategy_arguments={'folds': 5},
)
automl.fit(X_train, Y_train, metric=kappas)
# automl.refit(X_train, Y_train)
# Getting the rank of models
scores.save_log(automl, "log_imbal.txt")
# Predicting the model
scores.predict_and_save(automl,
X_test,
Y_test,
verbose=True,
file="prediction.txt")
# Save the model into binary code
filename = str(j) + 'imbal_model.sav'
pickle.dump(automl, open(filename, 'wb'))
elif option == 2:
# Applying SMOTE + ENN
X_resampled, Y_resampled = smoteenn.run(X_train,
Y_train,
smote_kind="svm")
resample = asc.AutoSklearnClassifier(
time_left_for_this_task=time + 10,
per_run_time_limit=time,
initial_configurations_via_metalearning=0)
resample.fit(X_resampled, Y_resampled, metric=kappas)
# Getting the rank of models
scores.save_log(resample, "log_bal.txt")
scores.predict_and_save(resample,
X_test,
Y_test,
verbose=True,
file="prediction_balanced.txt")
# Save the model into binary code
filename = str(j) + 'bal_model.sav'
pickle.dump(resample, open(filename, 'wb'))
else:
print("Invalid Option")
sys.exit()