outcome_col_name))
# Prepare data for classification
x_train = df_by_split['train'][feature_cols].values
y_train = np.ravel(df_by_split['train'][outcome_col_name])
x_test = df_by_split['test'][feature_cols].values
y_test = np.ravel(df_by_split['test'][outcome_col_name])
is_multiclass = len(np.unique(y_train)) > 2
# Perform hyper_searcher search
n_examples = x_train.shape[0]
n_features = x_train.shape[1]
splitter = Splitter(size=args.validation_size, random_state=args.random_seed, n_splits=args.n_splits, cols_to_group=args.key_cols_to_group_when_splitting)
key_train = splitter.make_groups_from_df(df_by_split['train'][key_cols])
# hyper_searcher search on validation over possible threshold values
# Make sure all candidates at least provide
# one instance of each class (positive and negative)
yproba_class1_vals = list()
for tr_inds, va_inds in splitter.split(x_train, groups=key_train):
x_valid = x_train[va_inds]
yproba_valid = x_valid
yproba_class1_vals.extend(yproba_valid)
unique_yproba_vals = np.unique(yproba_class1_vals)
if unique_yproba_vals.shape[0] == 1:
nontrivial_thr_vals = unique_yproba_vals
else:
# Try all thr values that would give at least one pos and one neg decision
df_by_split[split_name] = cur_df
outcome_col_name = args.outcome_col_name
# Prepare data for classification
x_train = df_by_split['train'][feature_cols].values.astype(np.float32)
y_train = np.ravel(df_by_split['train'][outcome_col_name])
x_test = df_by_split['test'][feature_cols].values.astype(np.float32)
y_test = np.ravel(df_by_split['test'][outcome_col_name])
if args.valid_csv_files is None:
# get the validation set
splitter = Splitter(
size=args.validation_size,
random_state=41,
n_splits=args.n_splits,
cols_to_group=args.key_cols_to_group_when_splitting)
# Assign training instances to splits by provided keys
key_train = splitter.make_groups_from_df(
df_by_split['train'][key_cols])
# get the train and validation splits
for ss, (tr_inds, va_inds) in enumerate(
splitter.split(x_train, y_train, groups=key_train)):
x_tr = x_train[tr_inds].copy()
y_tr = y_train[tr_inds].copy()
x_valid = x_train[va_inds]
y_valid = y_train[va_inds]
x_train = x_tr