7. Interpretation

import platform

import tensorflow as tf
tf.compat.v1.disable_v2_behavior()

import matplotlib
import matplotlib.pyplot as plt
plt.rcParams["font.family"] = "Times New Roman"

import numpy as np
import pandas as pd

import shap
from shap import Explanation
from shap.plots import beeswarm, violin, heatmap, waterfall
from shap import DeepExplainer, GradientExplainer, KernelExplainer

from umap import UMAP
from sklearn.manifold import TSNE
from sklearn.preprocessing import OneHotEncoder

from easy_mpl import imshow, pie, bar_chart
from easy_mpl.utils import create_subplots
from ai4water.postprocessing import PermutationImportance
from ai4water.postprocessing import PartialDependencePlot

from alibi.utils import gen_category_map
from alibi.explainers import plot_pd_variance
from alibi.explainers import PartialDependenceVariance

from utils import get_dataset, get_fitted_model, evaluate_model, \
    box_violin, shap_scatter, DYE_TYPES, ADSORBENT_TYPES, make_data

CAT_FEATURES = ['Adsorbent', 'Dye']

print(shap.__version__)

0.41.0

dataset, adsorbent_enc, dye_enc = get_dataset(encoding="ohe")

X_train, y_train = dataset.training_data()

***** Training *****
input_x shape:  (1059, 74)
target shape:  (1059, 1)

X_test, y_test = dataset.test_data()

***** Test *****
input_x shape:  (455, 74)
target shape:  (455, 1)

feature_names = dataset.input_features[0:10] + ['Adsorbent'] + ['Dye']

model, _ = get_fitted_model(model_type='functional')

***** Training *****
input_x shape:  (1059, 74)
target shape:  (1059, 1)
dot plot of model could not be plotted due to ('You must install pydot (`pip install pydot`) and install graphviz (see instructions at https://graphviz.gitlab.io/download/) ', 'for plot_model/model_to_dot to work.')

test_p = model.predict(x=X_test)

            argument test is deprecated and will be removed in future. Please
            use 'predict_on_test_data' method instead.
`Model.state_updates` will be removed in a future version. This property should not be used in TensorFlow 2.0, as `updates` are applied automatically.

evaluate_model(y_test, test_p)

mse 1982.9941791309736
rmse 44.53082279871969
r2 0.9862583591464813
r2_score 0.9862559257978472
mape inf
mae 19.717024301749014

train_p = model.predict(x=X_train)
print(f"Average value on prediction on training data {train_p.mean()}")

Average value on prediction on training data 191.0270233154297

evaluate_model(y_train, train_p)

mse 1732.7186196474518
rmse 41.625936862098996
r2 0.9908758607537946
r2_score 0.9908200504005319
mape inf
mae 18.47922520598649

adsorbent_ohe = X_test[:, 10:58]
adsorbent_original = adsorbent_enc.inverse_transform(adsorbent_ohe)
print(adsorbent_ohe.shape, adsorbent_original.shape)

(455, 48) (455, 1)

dye_ohe = X_test[:, 58:]
dye_original = dye_enc.inverse_transform(dye_ohe)

print(dye_ohe.shape, dye_original.shape)

(455, 16) (455, 1)

x_test_original = np.column_stack((X_test[:, 0:10], adsorbent_original, dye_original))

Deep Explainer

exp = DeepExplainer(model._model, data=X_train)

sv = exp.shap_values(X_test)[0]

dye_sv = sv[:, 58:].sum(axis=1)

adsorbent_sv = sv[:, 10:58].sum(axis=1)

shap_values = np.column_stack((sv[:, 0:10], dye_sv, adsorbent_sv))

print(shap_values.shape)

keras is no longer supported, please use tf.keras instead.
Your TensorFlow version is newer than 2.4.0 and so graph support has been removed in eager mode and some static graphs may not be supported. See PR #1483 for discussion.
(455, 12)

sv_df = pd.DataFrame(shap_values, columns=feature_names)
fig, axes = create_subplots(shap_values.shape[1])
for ax, col in zip(axes.flat, sv_df.columns):
    box_violin(ax=ax, data=sv_df[col], palette="Set2")
    ax.set_xlabel(col)
plt.tight_layout()
plt.show()

imshow(shap_values, aspect="auto", colorbar=True,
       xticklabels=feature_names, show=False)
plt.tight_layout()
plt.show()

shap_values_exp = Explanation(
    shap_values,
    data=x_test_original,
    feature_names=feature_names
)

Surface Area

df = pd.DataFrame(x_test_original, columns=feature_names)


shap_values_dye_dec = Explanation(
    shap_values,
    data=df.values,
    feature_names=feature_names
)

shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'])

FixedFormatter should only be used together with FixedLocator

<AxesSubplot:xlabel='Surface Area', ylabel='SHAP value for Surface Area'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'],
              feature_wrt = df['Volume (L)'], cmap = 'RdBu')

<AxesSubplot:xlabel='Surface Area', ylabel='SHAP value for Surface Area'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'],
              feature_wrt = df['Adsorption Time (min)'], cmap = 'RdBu')

<AxesSubplot:xlabel='Surface Area', ylabel='SHAP value for Surface Area'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'],
              feature_wrt = df['Pore Volume'], cmap = 'RdBu')

<AxesSubplot:xlabel='Surface Area', ylabel='SHAP value for Surface Area'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'],
              feature_wrt = df['Adsorption Temperature'], cmap = 'RdBu')

<AxesSubplot:xlabel='Surface Area', ylabel='SHAP value for Surface Area'>

feature_wrt = df['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Surface Area'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Initial Concentration

# It is the initial concentration of pollutant (aka adsorbate)

shap_scatter(shap_values=shap_values_dye_dec[:, 'Initial Concentration'])

<AxesSubplot:xlabel='Initial Concentration', ylabel='SHAP value for Initial Concentration'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Initial Concentration'],
              feature_wrt = df['Pyrolysis Time (min)'], cmap = 'RdBu')

<AxesSubplot:xlabel='Initial Concentration', ylabel='SHAP value for Initial Concentration'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Initial Concentration'],
              feature_wrt = df['Surface Area'], cmap = 'RdBu')

<AxesSubplot:xlabel='Initial Concentration', ylabel='SHAP value for Initial Concentration'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Initial Concentration'],
              feature_wrt = df['Pore Volume'], cmap = 'RdBu')

<AxesSubplot:xlabel='Initial Concentration', ylabel='SHAP value for Initial Concentration'>

feature_wrt = df['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Initial Concentration'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False
              )
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_dye_dec[:, 'Initial Concentration'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Initial Concentration'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Pyrolysis Temperature

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'])

<AxesSubplot:xlabel='Pyrolysis Temperature', ylabel='SHAP value for Pyrolysis Temperature'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = df['Surface Area'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Temperature', ylabel='SHAP value for Pyrolysis Temperature'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = df['Solution pH'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Temperature', ylabel='SHAP value for Pyrolysis Temperature'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = df['Adsorbent Loading'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Temperature', ylabel='SHAP value for Pyrolysis Temperature'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = df['Volume (L)'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Temperature', ylabel='SHAP value for Pyrolysis Temperature'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = df['Pore Volume'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Temperature', ylabel='SHAP value for Pyrolysis Temperature'>

feature_wrt = df['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False
              )
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Temperature'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Pyrolysis Time (min)

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'])

<AxesSubplot:xlabel='Pyrolysis Time (min)', ylabel='SHAP value for Pyrolysis Time (min)'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = df['Pore Volume'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Time (min)', ylabel='SHAP value for Pyrolysis Time (min)'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = df['Surface Area'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Time (min)', ylabel='SHAP value for Pyrolysis Time (min)'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = df['Initial Concentration'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Time (min)', ylabel='SHAP value for Pyrolysis Time (min)'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = df['Solution pH'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Time (min)', ylabel='SHAP value for Pyrolysis Time (min)'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = df['Pyrolysis Temperature'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pyrolysis Time (min)', ylabel='SHAP value for Pyrolysis Time (min)'>

feature_wrt = df['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False
              )
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Pyrolysis Time (min)'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Adsorption Time (min)

df_sv = pd.DataFrame(shap_values, columns=feature_names)
df_sv['adsorption_time_inp'] = x_test_original[:, 0]
df_sv = df_sv.loc[df_sv['adsorption_time_inp']<1399.0]
shap_values_ads = df_sv.iloc[:, 0:-1]

df_ads_t = pd.DataFrame(x_test_original, columns=feature_names)
df_ads_t = df_ads_t.loc[df_ads_t['Adsorption Time (min)']< 1399.0]


shap_values_exp_ads = Explanation(
    shap_values_ads.values.copy(),
    data=df_ads_t.values,
    feature_names=feature_names
)

shap_scatter(shap_values=shap_values_exp_ads[:, 'Adsorption Time (min)'])

<AxesSubplot:xlabel='Adsorption Time (min)', ylabel='SHAP value for Adsorption Time (min)'>

shap_scatter(
    shap_values=shap_values_exp_ads[:, 'Adsorption Time (min)'],
    feature_wrt = df_ads_t["Pore Volume"], cmap = 'RdBu')

<AxesSubplot:xlabel='Adsorption Time (min)', ylabel='SHAP value for Adsorption Time (min)'>

shap_scatter(
    shap_values=shap_values_exp_ads[:, 'Adsorption Time (min)'],
    feature_wrt = df_ads_t['Pyrolysis Time (min)'], cmap = 'RdBu'
)

<AxesSubplot:xlabel='Adsorption Time (min)', ylabel='SHAP value for Adsorption Time (min)'>

feature_wrt = df_ads_t['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(
    shap_values=shap_values_exp_ads[:, 'Adsorption Time (min)'],
    feature_wrt = feature_wrt.map(d),
    is_categorical=True,
    show=False
)
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_exp_ads[:, 'Adsorption Time (min)'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df_ads_t['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_exp_ads[:, 'Adsorption Time (min)'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Adsorbent Loading

It represents the amount of material (adsorbent) in the system

shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'])

<AxesSubplot:xlabel='Adsorbent Loading', ylabel='SHAP value for Adsorbent Loading'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = df['Solution pH'], cmap = 'RdBu')

<AxesSubplot:xlabel='Adsorbent Loading', ylabel='SHAP value for Adsorbent Loading'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = df['Pyrolysis Temperature'], cmap = 'RdBu')

<AxesSubplot:xlabel='Adsorbent Loading', ylabel='SHAP value for Adsorbent Loading'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = df['Volume (L)'], cmap = 'RdBu')

<AxesSubplot:xlabel='Adsorbent Loading', ylabel='SHAP value for Adsorbent Loading'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = df['Surface Area'], cmap = 'RdBu')

<AxesSubplot:xlabel='Adsorbent Loading', ylabel='SHAP value for Adsorbent Loading'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = df['Pore Volume'], cmap = 'RdBu')

<AxesSubplot:xlabel='Adsorbent Loading', ylabel='SHAP value for Adsorbent Loading'>

feature_wrt = df['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False
              )
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Adsorbent Loading'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Pore Volume

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'])

<AxesSubplot:xlabel='Pore Volume', ylabel='SHAP value for Pore Volume'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'],
              feature_wrt = df['Surface Area'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pore Volume', ylabel='SHAP value for Pore Volume'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'],
              feature_wrt = df['Solution pH'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pore Volume', ylabel='SHAP value for Pore Volume'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'],
              feature_wrt = df['Volume (L)'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pore Volume', ylabel='SHAP value for Pore Volume'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'],
              feature_wrt = df['Pyrolysis Temperature'], cmap = 'RdBu')

<AxesSubplot:xlabel='Pore Volume', ylabel='SHAP value for Pore Volume'>

feature_wrt = df['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False,
              )
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Pore Volume'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Solution pH

shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'])

<AxesSubplot:xlabel='Solution pH', ylabel='SHAP value for Solution pH'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'],
              feature_wrt = df['Volume (L)'], cmap = 'RdBu')

<AxesSubplot:xlabel='Solution pH', ylabel='SHAP value for Solution pH'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'],
              feature_wrt = df['Surface Area'], cmap = 'RdBu')

<AxesSubplot:xlabel='Solution pH', ylabel='SHAP value for Solution pH'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'],
              feature_wrt = df['Pore Volume'], cmap = 'RdBu')

<AxesSubplot:xlabel='Solution pH', ylabel='SHAP value for Solution pH'>

shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'],
              feature_wrt = df['Pyrolysis Temperature'], cmap = 'RdBu')

<AxesSubplot:xlabel='Solution pH', ylabel='SHAP value for Solution pH'>

feature_wrt = df['Dye']
d = {k:DYE_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False,
              )
plt.tight_layout()
plt.show()

shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'],
              feature_wrt = feature_wrt,
              is_categorical=True,
             palette_name="tab20",
             show=False)
plt.tight_layout()
plt.show()

feature_wrt = df['Adsorbent']
d = {k:ADSORBENT_TYPES[k] for k in feature_wrt.unique()}
shap_scatter(shap_values=shap_values_dye_dec[:, 'Solution pH'],
              feature_wrt = feature_wrt.map(d),
              is_categorical=True,
             show=False)
plt.tight_layout()
plt.show()

Bar chart of shap values for three classes i.e., ‘Physical properties’, ‘Synthesis Conditions’ and ‘Adsorption Experimental Conditions

sv_bar = np.mean(np.abs(shap_values_exp.values), axis=0)

colors = {'Adsorption Experimental Conditions': '#60AB7B',
          'Physical Properties': '#F9B234',
          'Synthesis Conditions': '#E91B23'}

classes = ['Adsorption Experimental Conditions', 'Synthesis Conditions',
           'Synthesis Conditions', 'Adsorption Experimental Conditions',
           'Adsorption Experimental Conditions', 'Adsorption Experimental Conditions',
           'Adsorption Experimental Conditions', 'Adsorption Experimental Conditions',
           'Physical Properties', 'Physical Properties', 'Synthesis Conditions',
           'Adsorption Experimental Conditions']

df = pd.DataFrame({'features': shap_values_exp.feature_names, 'classes': classes})

ax = bar_chart(sv_bar, shap_values_exp.feature_names,
          bar_labels=sv_bar, bar_label_kws={'label_type':'edge',
                                            'fontsize': 10,
                                            'weight': 'bold'},
          show=False, sort=True, color=['#60AB7B', '#E91B23', '#E91B23',
                                        '#60AB7B', '#60AB7B', '#60AB7B',
                                        '#60AB7B', '#60AB7B', '#F9B234',
                                        '#F9B234', '#E91B23', '#60AB7B'])
ax.spines[['top', 'right']].set_visible(False)
ax.set_xlabel(xlabel='mean(|SHAP value|)', fontsize=14, weight='bold')
ax.set_xticklabels(ax.get_xticks().astype(int), size=12, weight='bold')
ax.set_yticklabels(ax.get_yticklabels(), size=12, weight='bold')

labels = df['classes'].unique()
handles = [plt.Rectangle((0,0),1,1, color=colors[l]) for l in labels]
plt.legend(handles, labels, loc='lower right')

plt.tight_layout()
plt.show()

FixedFormatter should only be used together with FixedLocator

sv_norm = sv_bar / np.sum(sv_bar)

synthesis = np.sum(sv_norm[[2,1,10]])
physical = np.sum(sv_norm[[8,9]])
experimental = np.sum(sv_norm[[3,0,5,11,7,6,4]])

ax = pie(fractions=[experimental, physical, synthesis], labels=['','',''],
            colors=colors.values(),
         textprops={"fontsize":18, "weight":"bold"}, show=False,)
plt.tight_layout()
plt.show()

findfont: Font family ['Times New Roman'] not found. Falling back to DejaVu Sans.

heatmap(shap_values_exp, show=False)
plt.tight_layout()
plt.show()

findfont: Font family ['Times New Roman'] not found. Falling back to DejaVu Sans.
findfont: Font family ['Times New Roman'] not found. Falling back to DejaVu Sans.

heatmap(shap_values_exp, feature_values=shap_values_exp.abs.max(0), show=False)
plt.tight_layout()
plt.show()

heatmap(shap_values_exp, instance_order=shap_values_exp.sum(1), show=False)
plt.tight_layout()
plt.show()

beeswarm(shap_values_exp, show=False)
plt.tight_layout()
plt.show()

violin(shap_values, feature_names=feature_names, show=False)
plt.tight_layout()
plt.show()

This figure shows the learning of our model for one sample having maximum test prediction. E[f(X)] is average prediction for the training data. f(X) is the actual prediction for this sample. The figure tells us the negative or positive role of the input features in predicting the target equal to f(X) starting from E[f(X)].

# Default SHAP colors
default_pos_color = "#ff0051"
default_neg_color = "#008bfb"
# Custom colors
positive_color = "#ca0020"
negative_color = "#92c5de"

index = test_p.argmax()
e = Explanation(
    shap_values[index],
    base_values=exp.expected_value[0],
    data=x_test_original[index],
    feature_names=feature_names
)

waterfall(e, show=False, max_display=20)
# Change the colormap of the artists
for fc in plt.gcf().get_children():
    for fcc in fc.get_children():
        if isinstance(fcc, matplotlib.patches.FancyArrow):
            if matplotlib.colors.to_hex(fcc.get_facecolor()) == default_pos_color:
                fcc.set_facecolor(positive_color)
            elif matplotlib.colors.to_hex(fcc.get_facecolor()) == default_neg_color:
                fcc.set_color(negative_color)
        elif isinstance(fcc, plt.Text):
            if matplotlib.colors.to_hex(fcc.get_color()) == default_pos_color:
                fcc.set_color(positive_color)
            elif matplotlib.colors.to_hex(fcc.get_color()) == default_neg_color:
                fcc.set_color(negative_color)
axes = plt.gca()
axes.figure.set_size_inches((8,6))
plt.tight_layout()
plt.show()

This is same figure as above but for a different sample i.e., the minimum value of prediction.

index = test_p.argmin()
e = Explanation(
    shap_values[index],
    base_values=exp.expected_value[0],
    data=x_test_original[index],
    feature_names=feature_names
)

waterfall(e, show=False, max_display=20)
# Change the colormap of the artists
for fc in plt.gcf().get_children():
    for fcc in fc.get_children():
        if isinstance(fcc, matplotlib.patches.FancyArrow):
            if matplotlib.colors.to_hex(fcc.get_facecolor()) == default_pos_color:
                fcc.set_facecolor(positive_color)
            elif matplotlib.colors.to_hex(fcc.get_facecolor()) == default_neg_color:
                fcc.set_color(negative_color)
        elif isinstance(fcc, plt.Text):
            if matplotlib.colors.to_hex(fcc.get_color()) == default_pos_color:
                fcc.set_color(positive_color)
            elif matplotlib.colors.to_hex(fcc.get_color()) == default_neg_color:
                fcc.set_color(negative_color)
axes = plt.gca()
axes.figure.set_size_inches((8,6))
plt.tight_layout()
plt.show()

tsne = TSNE(n_components=2, random_state=313)
sv_2D = tsne.fit_transform(shap_values)

s = plt.scatter(sv_2D[:, 0], sv_2D[:, 1], c=y_test.reshape(-1,), cmap="Spectral",
            s=5)
plt.gca().set_aspect('equal', 'datalim')
plt.colorbar(s)
plt.title('TSNE projection of shap values', fontsize=18)
plt.show()

The default initialization in TSNE will change from 'random' to 'pca' in 1.2.
The default learning rate in TSNE will change from 200.0 to 'auto' in 1.2.

sv_umap = UMAP(n_components=2, random_state=313).fit_transform(shap_values)
s = plt.scatter(sv_umap[:, 0], sv_umap[:, 1], c=y_test.reshape(-1,),
            s=5, cmap="Spectral")
plt.gca().set_aspect('equal', 'datalim')
cbar = plt.colorbar(s)
cbar.ax.set_ylabel('Predicted Adsorption Capacity', rotation=270)
plt.title('UMAP projection of shap values', fontsize=18)
plt.show()

s = plt.scatter(sv_umap[:, 0], sv_umap[:, 1], c=X_test[:,0],
            s=5, cmap="Spectral")
plt.gca().set_aspect('equal', 'datalim')
cbar = plt.colorbar(s)
cbar.ax.set_ylabel('Adsorption Time (min)', rotation=270, labelpad=14)
plt.title('UMAP projection of shap values', fontsize=18)
plt.show()

s = plt.scatter(sv_umap[:, 0], sv_umap[:, 1], c=X_test[:,1],
            s=5, cmap="Spectral")
plt.gca().set_aspect('equal', 'datalim')
cbar = plt.colorbar(s)
cbar.ax.set_ylabel('Pyrolysis Temperature', rotation=270, labelpad=14)
plt.title('UMAP projection of shap values', fontsize=18)
plt.show()

s = plt.scatter(sv_umap[:, 0], sv_umap[:, 1], c=X_test[:,2],
            s=5, cmap="Spectral")
plt.gca().set_aspect('equal', 'datalim')
cbar = plt.colorbar(s)
cbar.ax.set_ylabel('Pyrolysis Time (min)', rotation=270, labelpad=14)
plt.title('UMAP projection of shap values', fontsize=18)
plt.show()

s = plt.scatter(sv_umap[:, 0], sv_umap[:, 1], c=X_test[:,3],
            s=5, cmap="Spectral")
plt.gca().set_aspect('equal', 'datalim')
cbar = plt.colorbar(s)
cbar.ax.set_ylabel('Initial Concentration', rotation=270, labelpad=14)
plt.title('UMAP projection of shap values', fontsize=18)
plt.show()

s = plt.scatter(sv_umap[:, 0], sv_umap[:, 1], c=X_test[:,4],
            s=5, cmap="Spectral")
plt.gca().set_aspect('equal', 'datalim')
cbar = plt.colorbar(s)
cbar.ax.set_ylabel('Solution pH', rotation=270, labelpad=14)
plt.title('UMAP projection of shap values', fontsize=18)
plt.show()

Kernel Explainer

if platform.system()=='Windows':

    X_train_summary = shap.kmeans(X_train, 10)

    exp = KernelExplainer(model.predict, X_train_summary)

    sv = exp.shap_values(X_test)[0]

    dye_sv = sv[:, 58:].sum(axis=1)

    adsorbent_sv = sv[:, 10:58].sum(axis=1)

    shap_values = np.column_stack((sv[:, 0:10], dye_sv, adsorbent_sv))

    print(shap_values.shape)

    # %%

    shap_values_exp = Explanation(
        shap_values,
        data=x_test_original,
        feature_names=feature_names
    )

    # %%

    beeswarm(shap_values_exp, show=False)
    plt.tight_layout()
    plt.show()

Gradient Explainer

exp = GradientExplainer(model._model, data=[X_train])

sv = exp.shap_values(X_test)[0]

dye_sv = sv[:, 58:].sum(axis=1)

adsorbent_sv = sv[:, 10:58].sum(axis=1)

shap_values = np.column_stack((sv[:, 0:10], dye_sv, adsorbent_sv))

print(shap_values.shape)

(455, 12)

shap_values_exp = Explanation(
    shap_values,
    data=x_test_original,
    feature_names=feature_names
)

beeswarm(shap_values_exp, show=False)
plt.tight_layout()
plt.show()

partial dependence plots

Now we will calculate partial dependence plot (PDP) and Individual Component Elements (ICE) curves using PartialDependencePlot class of ai4water.

pdp = PartialDependencePlot(
    model.predict,
    X_train,
    num_points=20,
    feature_names=model.input_features,
    show=False,
    save=False
)

We calculate pdp plots only once and then plot them with different options

feature = [f for f in model.input_features if f.startswith("Dye")]
pdp_vals, ice_vals = pdp.calc_pdp_1dim(X_train, feature)

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

ax = pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train, feature,
                        pdp_line_kws={'color': 'darkcyan'})
ax.set_xticklabels(dye_enc.categories_[0])
ax.set_xlabel("Dye")
plt.show()

ax = pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train, feature, ice=False,
                        pdp_line_kws={'color': 'darkcyan'})
ax.set_xticklabels(dye_enc.categories_[0])
ax.set_xlabel("Dye")
plt.show()

Calculate pdp and ice for Adsorbent

feature = [f for f in model.input_features if f.startswith("Adsorbent")]
pdp_vals, ice_vals = pdp.calc_pdp_1dim(X_train, feature)

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

ax = pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train, feature,
                        pdp_line_kws={'color': 'darkcyan'})
ax.set_xticklabels(adsorbent_enc.categories_[0])
ax.set_xlabel("Adsorbent")
plt.show()

ax = pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train, feature,
                        ice=False, pdp_line_kws={'color': 'darkcyan'})
ax.set_xticklabels(adsorbent_enc.categories_[0])
ax.set_xlabel("Adsorbent")
plt.show()

pdp and ice for Surface Area

pdp_vals, ice_vals = pdp.calc_pdp_1dim(X_train, 'Surface area')

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train, 'Surface area',
                   pdp_line_kws={'color': 'darkcyan'})
plt.show()

ax = pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train,
                        'Surface area', ice=False,
                        pdp_line_kws={'color': 'darkcyan'})
plt.tight_layout()
plt.show()

This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.

pdp and ice for Initial Concentration

feature = 'initial concentration'
pdp_vals, ice_vals = pdp.calc_pdp_1dim(X_train, feature)

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train, feature,
                   pdp_line_kws={'color': 'darkcyan'})
plt.tight_layout()
plt.show()

This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.

pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train,
                        feature, ice=False,
                        pdp_line_kws={'color': 'darkcyan'})
plt.tight_layout()
plt.show()

This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.

pdp and ice for Pyrolysis Temperature

feature = 'calcination_temperature'
pdp_vals, ice_vals = pdp.calc_pdp_1dim(X_train, feature)

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train, feature,
                   pdp_line_kws={'color': 'darkcyan'})
plt.tight_layout()
plt.show()

This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.

pdp._plot_pdp_1dim(pdp_vals, ice_vals, X_train,
                        feature, ice=False,
                        pdp_line_kws={'color': 'darkcyan'})
plt.tight_layout()
plt.show()

This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.

Accumulated Local Effects

from alepython import ale_plot

class Model:
    def predict(self, X):
        return model.predict(X).reshape(-1,)

ale_plot(train_set=pd.DataFrame(X_train, columns=model.input_features),
             model=Model(),
                  features=["Surface area"]
             )

            argument test is deprecated and will be removed in future. Please
            use 'predict_on_test_data' method instead.

<AxesSubplot:title={'center':"First-order ALE of feature 'Surface area'\nBins : 9 - Monte-Carlo : False"}, xlabel="Feature 'Surface area'">

ae_eff = ale_plot(train_set=pd.DataFrame(X_train, columns=model.input_features),
             model=Model(),
                  features=['calcination_temperature']
             )

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

ale_plot(train_set=pd.DataFrame(X_train, columns=model.input_features),
             model=Model(),
                  features=['initial concentration']
             )

            argument test is deprecated and will be removed in future. Please
            use 'predict_on_test_data' method instead.

<AxesSubplot:title={'center':"First-order ALE of feature 'initial concentration'\nBins : 8 - Monte-Carlo : False"}, xlabel="Feature 'initial concentration'">

ale_plot(train_set=pd.DataFrame(X_train, columns=model.input_features),
             model=Model(),
                  features=["Surface area", 'Pore volume']
             )

            argument test is deprecated and will be removed in future. Please
            use 'predict_on_test_data' method instead.

<AxesSubplot:title={'center':"Second-order ALE of features 'Surface area' and 'Pore volume'\nBins : 9x9"}, xlabel="Feature 'Surface area'", ylabel="Feature 'Pore volume'">

Permutation importance

Permutation importance quantifies reduction in model performance when we corrupt one feature column intentionally. The corruption in one feature column is carried out by randomly permuting its values n number of times. Then the average reduction in model performance is recorded as permutation feature importance for the feature.

cat_map = {'Catalyst': list(range(10, 58)), 'Anions': list(range(58, 74))}

pimp = PermutationImportance(
    model.predict, X_train, y_train,
    show=False,
    save=False,
    cat_map=cat_map,
    feature_names = model.input_features,
    n_repeats=20)

pimp.plot_1d_pimp()
plt.tight_layout()
plt.show()

pimp.plot_1d_pimp("barchart")
plt.tight_layout()
plt.show()

Partial Dependence Variance

We give the data to PartialDependenceVariance class label-encoded data. When is then one-hot-encoded inside the predictor_fn. Then the data is given to model to make a prediction

# first get data without any encoding
data, _, _ = make_data()
ads_ohe_encoder = OneHotEncoder(sparse=False)
ads_ohe_encoder.fit(data.loc[:, 'Adsorbent'].values.reshape(-1,1))

dye_ohe_encoder = OneHotEncoder(sparse=False)
dye_ohe_encoder.fit(data.loc[:, 'Dye'].values.reshape(-1,1))

# now get the label-encoded data. This will be passed to ``PartialDependenceVariance``
# class.
data_le, ads_le_encoder, dye_le_encoder = make_data(encoding="le")

def predictor_fn(X):

    # The X is given/suggested by ``PartialDependenceVariance`` which
    # means it is label-encoded. First inverse transform
    # to get the string columns
    ads_encoded = X[:, -2]
    ads_decoded = ads_le_encoder.inverse_transform(ads_encoded.astype(np.int16))
    ads_ohe_encoded = ads_ohe_encoder.transform(ads_decoded.reshape(-1,1))
    ads_cols = [f'Adsorbent_{i}' for i in range(ads_ohe_encoded.shape[1])]

    dye_encoded = X[:, -1]
    dye_decoded = dye_le_encoder.inverse_transform(dye_encoded.astype(np.int16))
    dye_ohe_encoded = dye_ohe_encoder.transform(dye_decoded.reshape(-1,1))
    dye_cols = [f'Dye_{i}' for i in range(dye_ohe_encoded.shape[1])]

    X = pd.DataFrame(X, columns=data.columns.tolist()[0:-1])
    X.pop('Adsorbent')
    X.pop('Dye')

    X[ads_cols] = ads_ohe_encoded
    X[dye_cols] = dye_ohe_encoded

    return model.predict(X.values).reshape(-1,)


category_map = gen_category_map(data)

pd_variance = PartialDependenceVariance(predictor=predictor_fn,
                                        feature_names=data.columns.tolist()[0:-1],
                                        categorical_names=category_map,
                                        target_names=["Adsorption"])

exp_importance = pd_variance.explain(X=data_le.values[:, 0:-1], method='importance')

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

argument test is deprecated and will be removed in future. Please
use 'predict_on_test_data' method instead.

bar_chart(
    exp_importance.feature_importance.reshape(-1,),
    sort=True,
    labels=data.columns.tolist()[0:-1],
    ax_kws=dict(tight_layout=True)
)

<AxesSubplot:>

f, ax = plt.subplots(3,4, figsize=(10,10))
plot_pd_variance(exp=exp_importance, summarise=False, ax=ax)
plt.subplots_adjust(hspace=0.1)
plt.show()