Generate candlestick plot from PmmlShapExplainerTest output

shap_value_candlestick_plot.py

import numpy as np
import matplotlib.pyplot as plt
import matplotlib as mpl

mpl.rcParams.update({
    "axes.facecolor": "DDDDDD",
    "axes.edgecolor": "white",
    "axes.grid": True,
    "axes.axisbelow": True,
    "legend.facecolor": "ffffff",
    "grid.color": "white",
    "grid.linestyle": "-",
    "savefig.bbox": "tight" 
})

feature_map = {
    "AGE":"Age",
    "AMT_INCOME_TOTAL":"Income",
    "CNT_CHILDREN":"# Children",
    "DAYS_EMPLOYED":"Days Employed",
    "FLAG_OWN_REALTY": "Own Realty?",
    "FLAG_WORK_PHONE": "Work Phone?",
    "FLAG_OWN_CAR": "Own Car?"
}

# vvvvvvvvvvvvvvvvvvv copy and paste output from PmmlShapExplainerTest here  vvvvvvvvvvvvvvvv
background_feature_means = [44.55, 74340.0, 2.01, 2478.0, 0.54, 0.46, 0.39] 
features = {
	"AGE": 57.000000,
	"AMT_INCOME_TOTAL": 111000.000000,
	"CNT_CHILDREN": 4.000000,
	"DAYS_EMPLOYED": 3700.000000,
	"FLAG_OWN_REALTY": 1.000000,
	"FLAG_WORK_PHONE": 0.000000,
	"FLAG_OWN_CAR": 1.000000,
}
prediction = 0.658974 
shap_values = {
	"AGE": (0.058538, 0.002538),
	"AMT_INCOME_TOTAL": (0.110987, 0.002538),
	"CNT_CHILDREN": (-0.019712, 0.002538),
	"DAYS_EMPLOYED": (0.049493, 0.002538),
	"FLAG_OWN_REALTY": (-0.002034, 0.002538),
	"FLAG_WORK_PHONE": (0.002363, 0.002538),
	"FLAG_OWN_CAR": (-0.006651, 0.006216),
}
fnull=[0.46598961291984575] 

# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

# plot setup
fig = plt.figure(figsize=(9,9),dpi=200)
ax = plt.gca()
pos = fnull[0]
barheight=.41
bar_colors = []
upwards = max(prediction, fnull[0])
downwards = min(prediction, fnull[0])
xlabels = []
colors = ['#ee0000', "#316DC1"]

# plot each shap value
for i,(k, (sv, conf)) in enumerate(shap_values.items()):
    # plot bar
    bar = ax.bar(i, height=sv, width=.84, bottom=pos, label="{}={:.0f}".format(feature_map[k], features[k]), color='#ee0000' if sv<0 else "#316DC1")
    bar_colors.append(bar.patches[0].get_facecolor())

    # place shap value onto plot
    if sv > 0:  
        ax.text(i, sv+pos+.001,  "+{:.2f}%".format(sv*100), fontsize=8, fontweight='bold', color=bar_colors[-1], horizontalalignment='center')
    else:
        ax.text(i, sv+pos-.001, "{:.2f}%".format(sv*100), fontsize=8, fontweight='bold', color=bar_colors[-1], horizontalalignment='center',verticalalignment='top')
    xlabels.append("{}\n={}".format(feature_map[k], features[k]))
    pos+=sv

# place joints across bars at top/bottom for clarity
pos = fnull[0]  
for i,(k, (sv, conf)) in enumerate(list(shap_values.items())[:-1]):   
    ax.plot((i-barheight, i+.5),(pos+sv, pos+sv), color=bar_colors[i])
    ax.plot((i+.5, i+1+barheight), (pos+sv, pos+sv), color=bar_colors[i+1])
    pos+=sv

# place horizontal line at prediction
ax.axhline(prediction, linewidth=1, color='k',alpha=.25)
ax.text(-1+.1, prediction+.001, "Prediction={:.2f}%".format(prediction*100), fontsize=10, horizontalalignment='left')

# place horizontal line at background value
ax.axhline(fnull[0], linewidth=1, color='k',alpha=.25)
ax.text(7-.1, fnull[0]+.001, "Background={:.2f}%".format(fnull[0]*100), fontsize=10,  horizontalalignment="right")

# label plot
plt.title("Per-Feature SHAP Values, Loan Approval")
plt.ylabel("Probability of Loan Approval")
b, t = .45, .7
plt.ylim(b,t)
plt.xlim(-1,7)
xticks = np.linspace(b,t,11)
plt.yticks(xticks, ["{:.1f}%".format(x*100) for x in xticks], fontsize=8)
plt.xticks(np.arange(0,7),xlabels, fontsize=8)
for ticklabel, (sv,_) in zip(plt.gca().get_xticklabels(), list(shap_values.values())):
    ticklabel.set_color(colors[0] if sv<0 else colors[1])
    
# plot and save
plt.savefig("loan_shap_values")
plt.show()