def process_ordinal_input(data,
key,
drop_percent=None,
drop_val=np.nan,
nac=False,
_print=print):
'''Helper function to perform processing on ordinal input,
where note this definition of ordinal means categorical ordinal...
so converting input to ordinal, but from categorical!
Parameters
----------
data : pandas DataFrame
BPt formatted df. Must contain a column with `key`
key : str
Column key of the column to process within `data` input.
drop_percent : float
% to drop
drop_val : NaN or int
If a row needs to be dropped, replace with drop_val
(default = np.nan)
Returns
----------
pandas DataFrame
The post-processed BPt formatted input df.
sklearn LabelEncoder
The sklearn labelencoder object mapping input to
transformed ordinal label
'''
# If requested, perform categorical dropping by percent
if drop_percent:
data = cat_drop_by_percent(data, key, drop_percent, drop_val, _print)
# Work on only non_dropped data / non NaN data if applicable
non_drop_data, non_drop_subjects =\
get_non_drop(data, key, drop_val)
# Encode ordinally
label_encoder = LabelEncoder()
non_drop_data[key] = label_encoder.fit_transform(non_drop_data[key])
non_drop_data[key] = non_drop_data[key].astype('category')
# Re-create data
data = put_non_drop_back(data, non_drop_subjects, non_drop_data)
# Check if nan was encoded if nac
if nac:
try:
nan_val = label_encoder.transform(['nan'])[0]
label_encoder.nan_val = nan_val
# Still need a NaN value, even if not already set
except ValueError:
label_encoder.classes_ =\
np.array(list(label_encoder.classes_) + ['nan'])
label_encoder.nan_val = len(label_encoder.classes_) - 1
return data, label_encoder
