def pred_target_plot(clf,X,Y,features_to_plot,label,grid_range=None):
from pdpbox import info_plots
figs = list()
axs = list()
pd.options.mode.chained_assignment = None # Turn warning msg off
# Extract the classifier object from the clf multilearn object
index = Y.columns.to_list().index(label)
clf = clf.classifiers_[index]
clf.verbose = False #Turn verbose off after this to tidy prints
for feature in features_to_plot:
if(grid_range is None):
fig, ax, summary_df = info_plots.actual_plot(clf,X,feature=feature,feature_name=feature,grid_type='equal')
else:
fig, ax, summary_df = info_plots.actual_plot(clf,X,feature=feature,feature_name=feature,grid_type='equal',
show_outliers='True',grid_range=grid_range)
figs.append(fig)
axs.append(ax)
pd.options.mode.chained_assignment = 'warn' # Turn warning msg back on
clf.verbose = True # reset
return figs, axs
def show_ICE_actual(self, features=[], feature_names=[]):
if len(features) != len(feature_names):
print("features and feature names must have same size")
return
for f,n in zip(features,feature_names):
info_plots.actual_plot(self.model,self.X_train,
feature=f,feature_name=n,predict_kwds={})
plt.xticks(rotation=90)
plt.show()
def predictionDistribution(data, pr, featureToExamine):
fig, axes, summary_df = info_plots.actual_plot(
model=pr,
X=data,
feature=featureToExamine,
feature_name=featureToExamine,
predict_kwds={})
save("predictionDistribution", fig=fig, plt=plt)
def info_actual_plot(self, feature, sample = 10000, predict_kwds = {}, which_classes=None, **kargs):
fig, axes, result = info_plots.actual_plot(
model=self.md,
X=self.sample(sample),
feature=feature, feature_name=feature,
predict_kwds=predict_kwds, which_classes = which_classes, **kargs)
self.info_actual_data = ResultDF(result, 'count')
plt.show()
def ice_pred_plot(self, feature, feature_name='Feature'):
'''
partial dependence plot - pdpbox version
https://towardsdatascience.com/introducing-pdpbox-2aa820afd312
'''
fig, axes, summary_df = info_plots.actual_plot(
model=self.model, X=self.X_train, feature=feature, feature_name=feature_name)
plt.plot()
def actual_plot(model,
X,
feature,
feature_name,
num_grid_points=10,
xticklabels=None,
show_percentile=False):
"""Wrapper for info_plots.actual_plot."""
fig, axes, summary_df = info_plots.actual_plot(
model=model,
X=X,
feature=feature,
feature_name=feature_name,
num_grid_points=num_grid_points,
show_percentile=show_percentile,
predict_kwds={},
)
if xticklabels is not None:
_ = axes["bar_ax"].set_xticklabels(xticklabels)
return fig, summary_df
# %% target_plot -------------------------------------------------------------
fig, axes, summary_df = info_plots.target_plot(
df=XY,
feature=x_cols[2],
feature_name=x_cols[2],
target=y_cols[0],
)
# %% actual_plot -------------------------------------------------------------
fig, axes, df = info_plots.actual_plot(
model=model,
X=X,
feature=x_cols[1],
feature_name=x_cols[1],
which_classes=[0, 3, 6],
predict_kwds={}, # !This should be passed to avoid a strange TypeError
)
# %% pdp_isolate: Preset -----------------------------------------------------
pdp_isolated_tmp = pdp.pdp_isolate(
model=model,
dataset=X,
model_features=x_cols,
feature=x_cols[0],
n_jobs=1,
)
# %% pdp_plot
titanic_features = test_titanic['features']
titanic_model = test_titanic['xgb_model']
titanic_target = test_titanic['target']
#Let's start with the gender
#Survivors based on their sex
fig, axes, summary_df = info_plots.target_plot(
df=titanic_data, feature='Sex', feature_name='gender', target=titanic_target
)
_ = axes['bar_ax'].set_xticklabels(['Female', 'Male'])
display(summary_df)
#Chance of survival our model give based on gender
fig, axes, summary_df = info_plots.actual_plot(
model=titanic_model, X=titanic_data[titanic_features], feature='Sex', feature_name='gender'
)
display(summary_df)
#PDP for the genderfeature
pdp_sex = pdp.pdp_isolate(
model=titanic_model, dataset=titanic_data, model_features=titanic_features, feature='Sex'
)
fig, axes = pdp.pdp_plot(pdp_sex, 'Sex', plot_lines=True, frac_to_plot=0.5)
_ = axes['pdp_ax'].set_xticklabels(['Female', 'Male'])
#Let's go on with the PassengerClass feature
#Firstly, the statistics of survivors based on their PassengerClass
fig, axes, summary_df = info_plots.target_plot(
df=titanic_data, feature='Pclass', feature_name='Pclass', target=titanic_target, show_percentile=True
)
display(summary_df)
#PDP Plots : Target Plot
for _ in X.columns.tolist():
fig,axes,summary_df=info_plots.target_plot(
df=df,
feature=_,
feature_name=_,
target=target_variable
)
#PDP Plots : Actual Plot
for _ in X.columns.tolist():
fig,axes,summary_df=info_plots.actual_plot(
model=baseline,
X=df[X.columns],
feature=_,
feature_name=_,
predict_kwds={}
)
fig, axes, summary_df = info_plots.actual_plot_interact(
model=baseline, X=df[X.columns], features=interactions_2way, feature_names=interactions_2way
)
#PDP Plot : Grid Plot
interactions = pdp.pdp_interact(
model=baseline, dataset=df, model_features=X.columns, features=interactions_2way
)
fig, axes = pdp.pdp_interact_plot(interactions,interactions_2way , plot_type='grid', x_quantile=True, plot_pdp=False)
# region Predictions boxplots
#Variables with high importance in the predicton:
# Internet service/Monthly charges
#Lenght of contract
#Tenure
#Online security
#Tech support
#defining a small function to unscale the x axis
def unscaling(x_scaled,X_col_unsc):
return ( np.round((x_scaled* X_col_unsc.std(axis=0)) + X_col_unsc.mean(axis=0),1).astype(int) )
# region Tenure
#Actual plot
fig, axes, summary_df =info_plots.actual_plot(model=best_rf.named_steps["classifier"], X=X_train,
feature="tenure",feature_name="Tenure",predict_kwds={},num_grid_points=11)
axes["bar_ax"].set_xticklabels(pd.qcut(X_train_ori.tenure,10,precision=3).values.categories.values)
#plt.savefig('Predi_boxplot_tenure.png', bbox_inches='tight')
#pickle.dump(fig,open("Predi_boxplot_tenure.pickle","wb"))
# endregion
# region Internet
fig, axes, summary_df =info_plots.actual_plot(model=best_rf.named_steps["classifier"], X=X_train,
feature=["InternetService_DSL","InternetService_Fiber optic","InternetService_No"],feature_name="Internet service",predict_kwds={},
base_features = X_train.columns.values.tolist()
feature_name = 'incident_severity'
pdp = pdp.pdp_isolate(model=model, dataset=X_test,
model_features = base_features,
feature = feature_name)
pdp.pdp_plot(pdp, feature_name)
plt.show()
fig, axes, summary_df = info_plots.target_plot(\
df=data, feature='capital-gains',
feature_name='capital-gains', target='fraud_reported',
show_percentile=True)
fig, axes, summary_df = info_plots.actual_plot(\
model, X_test, feature='capital-gains',
feature_name='capital-gains',predict_kwds={})
df.auto_model.unique()
fig, axes, summary_df = info_plots.target_plot(\
df=data, feature='auto_model',
feature_name='auto_model', target='fraud_reported',
show_percentile=True)
fig, axes, summary_df = info_plots.target_plot(\
df=data, feature='auto_make',
feature_name='auto_make', target='fraud_reported',
show_percentile=True)
fig, axes, summary_df = info_plots.target_plot(\