In this version I have:
- Fixed some potential bugs in
split_bins
,ks_table_plot
,ks_psi_plot
,ks_value
. - Add a new function
pred_score
for predicting new data using trained models. - Provide new functions
perf_table
,roc_plot
,ks_plot
,lift_plot
,psi_plot
for model validation drawings.
In this version I have:
- Fixed some potential bugs in
get_names
,digits_num
In this version I have:
- Add a function
data_exploration
for data exploration. - Fixed some potential bugs in
missing_proc
,outliers_proc
,get_names
- In
lasso_filter
,AUC
&K-S
is added to select the best lambda. In this way, not only can the set of variables that makes the AUC or K-S maximized be selected, but also the multicollinearity (which is difficult to eliminate by AIC in stepwise regression), can be minimized. That means instead of stepwise regression, the optimal combination of variables can be selected by lasso to solve the regression problem. - Visualize the number of variables,
K-S
orAUC
values corresponding to different lambda. - Provide new functions``auc_value
\
ks_value`, which can calculate Kolmogorov-Smirnov (K-S) & AUC of multiple model results quickly.