def fit(self, data):
if data.kfold > 1:
cv_eval = {}
for k, cv_fold in enumerate(data.Xy_train.keys()):
# print(' cv_fold: ', cv_fold)
[(X_train, y_train), (X_val, y_val)] = data.Xy_train[cv_fold]
X_train, X_val = from_2d_array_to_nested(X_train), from_2d_array_to_nested(X_val)
knn = KNeighborsTimeSeriesClassifier(n_neighbors=5, distance="dtw", n_jobs=-1)
knn.fit(X_train, y_train)
eval_metrics = weareval.eval_output(knn.predict(X_val), y_val, tasktype=data.tasktype)
cv_eval[cv_fold] = {'model': knn,
# 'data': [(X_train, y_train), (X_val, y_val)], # store just IDs?
'metric': eval_metrics['mae'] if data.tasktype=='regression' else eval_metrics['balanced_acc_adj'],
'metrics': eval_metrics}
# retain only best model
tmp = {cv_fold:cv_eval[cv_fold]['metric'] for cv_fold in cv_eval.keys()}
bst_fold = min(tmp, key=tmp.get) if data.tasktype=='regression' else max(tmp, key=tmp.get)
self.knn = cv_eval[bst_fold]['model']
return {'model': self.knn, 'metrics': cv_eval[bst_fold]['metrics']}
else:
X_train, y_train = data.Xy_train
X_val, y_val = data.Xy_val
X_train, X_val = from_2d_array_to_nested(X_train), from_2d_array_to_nested(X_val)
self.knn = knn = KNeighborsTimeSeriesClassifier(n_neighbors=5, distance="dtw", n_jobs=-1)
self.knn.fit(X_train, y_train)
eval_metrics = weareval.eval_output(self.knn.predict(X_val), y_val, tasktype=data.tasktype)
return {'model': self.knn, 'metrics': eval_metrics}
def fit(self, data):
if data.kfold > 1:
cv_eval = {}
for k, cv_fold in enumerate(data.Xy_train.keys()):
[(X_train, y_train), (X_val, y_val)] = data.Xy_train[cv_fold]
X_train, X_val = from_2d_array_to_nested(X_train), from_2d_array_to_nested(X_val)
tsf = ComposableTimeSeriesForestRegressor(
n_jobs=-1) if data.tasktype=='regression' else ComposableTimeSeriesForestClassifier(
n_jobs=-1)
tsf.fit(X_train, y_train)
eval_metrics = weareval.eval_output(tsf.predict(X_val), y_val, tasktype=data.tasktype)
cv_eval[cv_fold] = {'model': tsf,
# 'data': [(X_train, y_train), (X_val, y_val)], # store just IDs?
'metric': eval_metrics['mae'] if data.tasktype=='regression' else eval_metrics['balanced_acc_adj'],
'metrics': eval_metrics}
# retain only best model
tmp = {cv_fold:cv_eval[cv_fold]['metric'] for cv_fold in cv_eval.keys()}
bst_fold = min(tmp, key=tmp.get) if data.tasktype=='regression' else max(tmp, key=tmp.get)
self.tsf = cv_eval[bst_fold]['model']
return {'model': self.tsf, 'metrics': cv_eval[bst_fold]['metrics']}
else:
X_train, y_train = data.Xy_train
X_val, y_val = data.Xy_val
X_train, X_val = from_2d_array_to_nested(X_train), from_2d_array_to_nested(X_val)
self.tsf = ComposableTimeSeriesForestRegressor(
n_jobs=-1) if data.tasktype=='regression' else ComposableTimeSeriesForestClassifier(
n_jobs=-1)
self.tsf.fit(X_train, y_train)
eval_metrics = weareval.eval_output(self.tsf.predict(X_val), y_val, tasktype=data.tasktype)
return {'model': self.tsf, 'metrics': eval_metrics}
def fit(self, target_name, smote=True, verbose=False):
X, y = self.data.to_numpy(
dtype=np.float32), self.get_model_y(target_name)
# CV splits
for i, (train_idx,
test_idx) in enumerate(self.splitter.split(X, y, self.grps)):
X_train, y_train = X[train_idx], y[train_idx]
X_test, y_test = X[test_idx], y[test_idx]
if 'cat' in self.voi[target_name]: # SMOTE
fpr, tpr = dict(), dict()
au_roc, eval_metrics = dict(), dict()
for jj in range(y_train.shape[1]):
if smote:
oversample = SMOTE(k_neighbors=3)
# print(f'kk: {kk}\tX_train: {X_train.shape}\ty_train: {y_train.shape}')
try:
X_train_mod, y_train_mod = oversample.fit_resample(
X_train, y_train[:, jj])
except ValueError:
print(
"\n{}-th class cannot be computed. Too few n_samples. Skipping"
.format(jj))
if verbose:
print('{} class frequencies:'.format(catvar))
for jjj in range(y_train.shape[1]):
print(
f"j: {jjj}\t0: {(y_train[:, jjj]==0).sum()}\t1: {(y_train[:, jjj]==1).sum()}"
)
continue
del oversample
else:
X_train_mod, y_train_mod = X_train, y_train
# model/eval
model = KNeighborsClassifier(n_jobs=12)
model.fit(X_train_mod, y_train_mod)
(fpr[jj], tpr[jj], thresholds) = sklmetrics.roc_curve(
y_test[:, jj],
model.predict_proba(X_test)[:, 1])
au_roc[jj] = sklmetrics.auc(fpr[jj], tpr[jj])
eval_metrics[jj] = weareval.eval_output(
model.predict(X_test),
y_test[:, jj],
tasktype='regression' if 'cont'
in self.voi[target_name] else 'classification')
# scores = cross_val_score(lr, X_train_mod, y_train_mod, cv=5, scoring='roc_curve')
self.res[f'{target_name}_fold{i}'] = (fpr, tpr, au_roc,
eval_metrics
) # (lr, scores)
else:
model = KNeighborsRegressor(n_jobs=12)
model.fit(X_train, y_train)
self.res[f'{target_name}_fold{i}'] = weareval.eval_output(
model.predict(X_test),
y_test,
tasktype='regression'
if 'cont' in self.voi[target_name] else 'classification')
def eval_test(self, data):
X_test, y_test = data.Xy_test
if self.downsample_inference:
idx = np.random.choice(np.arange(X_test.shape[0]), 20, replace=False)
X_test, y_test = X_test[idx], y_test[idx]
eval_metrics = weareval.eval_output(self.kNNDTW.predict(X_test), y_test, tasktype=data.tasktype)
return eval_metrics
def fit(self, data):
if data.kfold > 1:
cv_eval = {}
for k, cv_fold in enumerate(data.Xy_train.keys()):
print(' starting kfold=', cv_fold)
[(X_train, y_train), (X_val, y_val)] = data.Xy_train[cv_fold]
kNNDTW = KNeighborsRegressor(
n_jobs=-1,
algorithm='ball_tree',
weights='distance',
metric=self.dtw) if data.tasktype=='regression' else KNeighborsClassifier(
n_jobs=-1,
algorithm='ball_tree',
weights='distance',
metric=self.dtw)
kNNDTW.fit(X_train, y_train)
if self.downsample_inference:
idx = np.random.choice(np.arange(X_val.shape[0]), 20, replace=False)
X_val, y_val = X_val[idx], y_val[idx]
eval_metrics = weareval.eval_output(kNNDTW.predict(X_val), y_val, tasktype=data.tasktype)
cv_eval[cv_fold] = {'model': kNNDTW,
# 'data': [(X_train, y_train), (X_val, y_val)], # store just IDs?
'metric': eval_metrics['mae'] if data.tasktype=='regression' else eval_metrics['balanced_acc_adj'],
'metrics': eval_metrics}
# retain only best model
tmp = {cv_fold:cv_eval[cv_fold]['metric'] for cv_fold in cv_eval.keys()}
bst_fold = min(tmp, key=tmp.get) if data.tasktype=='regression' else max(tmp, key=tmp.get)
self.kNNDTW = cv_eval[bst_fold]['model']
return {'model': self.kNNDTW, 'metrics': cv_eval[bst_fold]['metrics']}
else:
X_train, y_train = data.Xy_train
X_val, y_val = data.Xy_val
self.kNNDTW = kNNDTW = KNeighborsRegressor(
n_jobs=-1,
algorithm='ball_tree',
weights='distance',
metric=self.dtw) if data.tasktype=='regression' else KNeighborsClassifier(
n_jobs=-1,
algorithm='ball_tree',
weights='distance',
metric=self.dtw)
self.kNNDTW.fit(X_train, y_train)
if self.downsample_inference:
idx = np.random.choice(np.arange(X_val.shape[0]), 20, replace=False)
X_val, y_val = X_val[idx], y_val[idx]
eval_metrics = weareval.eval_output(self.kNNDTW.predict(X_val), y_val, tasktype=data.tasktype)
return {'model': self.kNNDTW, 'metrics': eval_metrics}
def fit(self, data):
if data.kfold > 1:
cv_eval = {}
for k, cv_fold in enumerate(data.Xy_train.keys()):
[(X_train, y_train), (X_val, y_val)] = data.Xy_train[cv_fold]
kNN = KNeighborsRegressor(
n_jobs=16
) if self.tasktype == 'regression' else KNeighborsClassifier(
n_jobs=16)
kNN.fit(X_train, y_train)
eval_metrics = weareval.eval_output(kNN.predict(X_val),
y_val,
tasktype=self.tasktype)
cv_eval[cv_fold] = {
'model':
kNN,
# 'data': [(X_train, y_train), (X_val, y_val)], # store just IDs?
'metric':
eval_metrics['mae'] if self.tasktype == 'regression' else
eval_metrics['balanced_acc_adj'],
'metrics':
eval_metrics
}
# retain only best model
tmp = {
cv_fold: cv_eval[cv_fold]['metric']
for cv_fold in cv_eval.keys()
}
bst_fold = min(
tmp, key=tmp.get) if self.tasktype == 'regression' else max(
tmp, key=tmp.get)
self.kNN = cv_eval[bst_fold]['model']
return {'model': self.kNN, 'metrics': cv_eval[bst_fold]['metrics']}
else:
X_train, y_train = data.Xy_train
X_val, y_val = data.Xy_val
self.kNN = KNeighborsRegressor(
n_jobs=12
) if self.tasktype == 'regression' else KNeighborsClassifier(
n_jobs=12)
self.kNN.fit(X_train, y_train)
eval_metrics = weareval.eval_output(self.kNN.predict(X_val),
y_val,
tasktype=self.tasktype)
return {'model': self.kNN, 'metrics': eval_metrics}
def fit(self, data):
params = {
'boosting_type': 'gbdt',
'verbosity': 0}
if data.tasktype == 'regression':
params['objective'] = 'regression',
else:
if len(data.Xy_test[1].shape) > 1:
params['objective'] = 'multiclass',
else:
params['objective'] = 'binary',
if data.kfold > 1:
cv_eval = {}
for k, cv_fold in enumerate(data.Xy_train.keys()):
[(X_train, y_train), (X_val, y_val)] = data.Xy_train[cv_fold]
lgb_train = lgb.Dataset(X_train, y_train)
lgb_eval = lgb.Dataset(X_val, y_val)
gbm = lgb.train(params,
lgb_train,
valid_sets=lgb_eval,
callbacks=[lgb.early_stopping(stopping_rounds=5)])
eval_metrics = weareval.eval_output(gbm.predict(X_val, num_iteration=gbm.best_iteration), y_val, tasktype=data.tasktype)
cv_eval[cv_fold] = {'model': gbm,
# 'data': [(X_train, y_train), (X_val, y_val)], # store just IDs?
'metric': eval_metrics['mae'] if data.tasktype=='regression' else eval_metrics['balanced_acc_adj'],
'metrics': eval_metrics}
# retain only best model
tmp = {cv_fold:cv_eval[cv_fold]['metric'] for cv_fold in cv_eval.keys()}
bst_fold = min(tmp, key=tmp.get) if data.tasktype=='regression' else max(tmp, key=tmp.get)
self.gbm = cv_eval[bst_fold]['model']
return {'model': self.gbm, 'metrics': cv_eval[bst_fold]['metrics']}
else:
X_train, y_train = data.Xy_train
X_val, y_val = data.Xy_val
lgb_train = lgb.Dataset(X_train, y_train)
lgb_eval = lgb.Dataset(X_val, y_val)
self.gbm = lgb.train(params,
lgb_train,
valid_sets=lgb_eval,
callbacks=[lgb.early_stopping(stopping_rounds=5)])
eval_metrics = weareval.eval_output(self.gbm.predict(X_val, num_iteration=gbm.best_iteration), y_val, tasktype=data.tasktype)
return {'model': self.gbm, 'metrics': eval_metrics}
def eval_test(self, data):
X_test, y_test = data.Xy_test
eval_metrics = weareval.eval_output(self.kNN.predict(X_test), y_test, tasktype=data.tasktype)
return eval_metrics
def eval_test(self, data):
X_test, y_test = data.Xy_test
X_test = from_2d_array_to_nested(X_test)
eval_metrics = weareval.eval_output(self.knn.predict(X_test), y_test, tasktype=data.tasktype)
return eval_metrics