05. Sepsis shap by timestep (boxplot)

This script visualizes SHAP (SHapley Additive exPlanations) values to analyze feature importance across different time periods. It loads a pre-computed shap.csv file obtained for sepsis patients and a LSTM predictive model and uses Plotly to generate a comprehensive boxplot.

The resulting plot displays the distribution of SHAP values on the y-axis for each feature at every timestep on the x-axis. This makes it easy to quickly identify which features have the most significant impact on the model’s predictions (those with SHAP values of higher magnitude) and to see how that influence changes over time.

# Libraries
import pandas as pd
import numpy as np
import matplotlib as mpl
import plotly.express as px

from plotly.io import show
from plotly.colors import n_colors
from plotly.express.colors import sample_colorscale

# See https://matplotlib.org/devdocs/users/explain/customizing.html
mpl.rcParams['axes.titlesize'] = 8
mpl.rcParams['axes.labelsize'] = 8
mpl.rcParams['xtick.labelsize'] = 8
mpl.rcParams['ytick.labelsize'] = 8
mpl.rcParams['legend.fontsize'] = 7
mpl.rcParams['legend.handlelength'] = 1
mpl.rcParams['legend.handleheight'] = 1
mpl.rcParams['legend.loc'] = 'upper left'

try:
    __file__
    TERMINAL = True
except:
    TERMINAL = False


# Load data
data = pd.read_csv('../../datasets/shap/shap.csv')

# Show
if TERMINAL:
    print("\nData:")
    print(data)
data

# Number of colors
N = data.features.nunique()

# see https://plotly.com/python/builtin-colorscales/#discrete-color-sequences
# see https://plotly.github.io/plotly.py-docs/generated/plotly.express.box.html

# generate an array of rainbow colors by fixing the saturation and lightness of the
# HSL representation of colour and marching around the hue. Plotly accepts any CSS
# color format, see e.g. http://www.w3schools.com/cssref/css_colors_legal.asp.
c0 = ['hsl('+str(h)+',50%'+',50%)'
    for h in np.linspace(0, 360, N)]

# More colors
x = np.linspace(0, 1, N)
c1 = sample_colorscale('viridis', list(x))
c2 = sample_colorscale('RdBu', list(x))
c3 = sample_colorscale('Jet', list(x))
c4 = sample_colorscale('Agsunset', list(x))

# .. note:: Remove width and size if running locally.

# Boxplot
fig = px.box(data, x='timestep', y='shap_values',
             color='features', color_discrete_sequence=c4,
             points='outliers', width=750, height=900)

# .. note:: If using widescreen, commenting the legend section
#           will automatically generate a vertical legend with
#           scrolling if needed. For display purposes in the
#           docs we have included the legend on top.

# Update layout
fig.update_layout(
    #margin={
    #    'l': 0,
    #    'r': 0,
    #    'b': 0,
    #    't': 0,
    #    'pad': 4
    #},
    legend=dict(
        orientation="h",
        entrywidth=140,
        yanchor="bottom",
        y=1.02,
        xanchor="right",
        x=1,
        #font=dict(
        #    family="Courier",
        #    size=7,
        #    #color="black"
        #),
    ),
    paper_bgcolor='rgba(0,0,0,0)',  # transparent
    plot_bgcolor='rgba(0,0,0,0)'  # transparent
)

# Update xaxis
fig.update_xaxes(
    mirror=False,
    ticks='outside',
    showline=False,
    linecolor='black',
    gridcolor='lightgrey'
)

# Update yaxis
fig.update_yaxes(
    mirror=False,
    ticks='outside',
    showline=False,
    linecolor='black',
    gridcolor='lightgrey'
)

# Show
show(fig)