def get_discretizer(x,
y,
continuous_features=None,
seed=None,
min_depth=0) -> MDLP:
discretizer = MDLP(random_state=seed, min_depth=min_depth)
if continuous_features is not None:
if continuous_features.dtype == np.bool:
continuous_features = np.arange(
len(continuous_features))[continuous_features]
discretizer.fit(x, y, continuous_features)
return discretizer
def num2cate_fit(df, min=2):
'''
Arg
df (Panda dataframes); the last col must be class, int 0 or 1
min (int): The minimum depth of the interval splitting. Overrides
the MDLP stopping criterion. If the entropy at a given interval
is found to be zero before `min_depth`, the algorithm will stop.
Return
mdlp (MDLP instance): transform, can be used to transform samples
'''
Y = df.iloc[:, -1].values
continuous_features =df.iloc[:, :-1].select_dtypes(include=['int64','float64']).columns.tolist()
continuous_features.sort() # ensoure the features order between fit and transform
X = df[continuous_features].values
mdlp = MDLP(min_depth=min)
mdlp.fit(X, Y) # X, Y should be numpy array
return mdlp
def get_raw_bins(column, target):
transformer = MDLP()
transformer = transformer.fit(column, target)
return list(transformer.cut_points_[0])
