def fit(self, tree, model_name):
self.initialize(model_name)
self.load()
self.log_params()
if model_name != 'DT':
model = tree(n_estimators=self.n_estimators,
max_depth=self.max_depth,
min_samples_split=self.min_samples_split,
min_samples_leaf=self.min_samples_leaf,
max_features=self.max_features,
bootstrap=self.bootstrap,
n_jobs=self.n_jobs,
verbose=self.verbose)
else:
model = tree(max_depth=self.max_depth,
min_samples_split=self.min_samples_split,
min_samples_leaf=self.min_samples_leaf,
max_features=self.max_features)
model.fit(self.X_train.values, self.Y_train.values)
print(f"{model_name} is fitted")
if self.should_cross_val:
scores = cross_val_score(model,
self.X,
self.Y,
cv=self.k,
verbose=0)
self.log.info(
f"---- Cross validation with {self.k} groups----\n\nThe results on each split"
+ str(scores) + "\n")
self.log.info(
f"The average of the cross validation is {np.mean(scores):.2f}\n"
)
print(
f"|- Cross validation is done for {model_name}. Accuracy: {np.mean(scores):.2f} -|"
)
evaluate_classification(
['OnTrain', self.X_train, self.Y_train, self.dates_train],
['OnTest', self.X_test, self.Y_test, self.dates_test],
direc=self.directory,
model=model,
model_name=model_name,
logger=self.log,
slicer=1)
joblib.dump(model, self.directory + f"/{model_name}.pkl")
plot_roc(
pd.get_dummies(self.Y_test, drop_first=False).values,
model.predict_proba(self.X_test), self.dl.classes_, self.directory)
# Plotting the Importances
report_feature_importance(self.directory, model.feature_importances_,
self.X, self.Y, self.n_top_features,
model_name, self.log)
def fit(self):
model = LogisticRegression(C = 100, fit_intercept = True, penalty= 'l1', solver = 'liblinear')
model.fit(self.X_train, self.Y_train)
self.dl.log.info("---Model Coeffs---\n" + str(model.coef_))
coeffs = {}
for i in range (len(model.coef_[0])):
coeffs[self.X.columns[i]] = model.coef_[0][i]
coeffs['c'] = model.intercept_[0]
evaluate_classification(['OnTrain', self.X_train, self.Y_train, self.dates_train],
['OnTest', self.X_test, self.Y_test, self.dates_test],
direc = self.directory,
model = model,
model_name = self.model_name,
logger = self.log,
slicer = 1)
plot_roc(pd.get_dummies(self.Y_test, drop_first = False).values,
model.predict_proba(self.X_test),
self.dl.classes_,
self.directory)
joblib.dump(model, self.directory + f"/Logit.pkl")
# Plotting the Importances
report_feature_importance(self.directory, model.coef_[0], self.X, self.Y,
self.n_top_features, "Logit", self.log)
def fit(self):
self.set_params()
self.log_params()
self.model = SVC(C=self.C, kernel=self.kernel, gamma=self.gamma)
self.model.fit(self.X_train, self.Y_train)
evaluate_classification(
['OnTrain', self.X_train, self.Y_train, self.dates_train],
['OnTest', self.X_test, self.Y_test, self.dates_test],
direc=self.directory,
model=self.model,
model_name=self.model_name,
logger=self.log,
slicer=1)
joblib.dump(self.model, self.directory + f"/{self.model_name}.pkl")
# Plotting the Importances
if self.kernel == 'linear':
report_feature_importance(self.directory, self.model.coef_[0],
self.X_train.columns,
self.n_top_features, self.model_name,
self.log)
def fit(self, n = 5):
self.log.info(f'KNN Classifier is about to be fit on {self.name} with n = {n}')
model = KNeighborsClassifier(n_neighbors=n, n_jobs = -1)
model.fit(self.X_train, self.Y_train)
evaluate_classification(['OnTrain', self.X_train, self.Y_train, self.dates_train],
['OnTest', self.X_test, self.Y_test, self.dates_test],
direc = self.directory,
model = model,
model_name = model_name,
logger = self.log,
slicer = 1)
def get_report(self):
self.load_model()
y_train_pred = self.predict_set(self.X_train)
y_test_pred = self.predict_set(self.X_test)
evaluate_classification(['OnTrain', self.X_train, self.Y_original_train, self.dates_train, y_train_pred],
['OnTest', self.X_test, self.Y_original_test, self.dates_test, y_test_pred],
direc = self.directory,
model = self.model,
model_name = self.model_name,
logger = self.log,
slicer = 1)
def fit(self):
value_counts = self.Y.value_counts()
value_counts = value_counts / value_counts.sum()
y_pred_train = np.random.choice(value_counts.index,
size=len(self.Y_train),
p=value_counts.values)
y_pred_test = np.random.choice(value_counts.index,
size=len(self.Y_test),
p=value_counts.values)
evaluate_classification([
'OnTrain', self.X_train, self.Y_train, self.dates_train,
y_pred_train
], ['OnTest', self.X_test, self.Y_test, self.dates_test, y_pred_test],
direc=self.directory,
model_name=self.model_name,
logger=self.log,
slicer=1)