svm_plot.py

import matplotlib.pyplot as plt
from netCDF4 import Dataset
import pandas as pd
import numpy as np
import scienceplots

# Import netCDF file
ncfile = './svm.nc'
df     = Dataset(ncfile).variables

# Prepare Data to Plot
x       = df['x'][:]
y       = df['y'][:]
g       = df['g'][:]
g_hat   = df['g_hat'][:]
w       = df['w'][:]
b       = df['b'][0]
f_hat   = df['f_hat'][:]
z       = df['z'][:]
tpr     = df['tpr'][:]
fpr     = df['fpr'][:]
auc     = df['auc'][0]

domain  = np.linspace(x.min(), x.max(), 1000)
hyper   = -w[0] * domain + b
platt   = z


# Plot params
pparam = dict(
    xlabel = r'$x$',
    ylabel = r'$y$',
    title = r"SVM",
    xscale = 'linear',
    yscale = 'linear',
    xlim = (-6, 6),
    ylim = (y.min(), y.max()),
)

# Plot
with plt.style.context(["science", "nature"]):
    fig, ax = plt.subplots()
    ax.autoscale(tight=True)
    ax.set(**pparam)
    ax.plot(domain, hyper, 'k--', label=r'$\hat{g}(x)$')
    ax.scatter(x, y, c=g, cmap='bwr', s=1, alpha=0.5, label=r'$g(x)$')
    ax.legend()
    fig.savefig('svm.png', dpi=300, bbox_inches='tight')

# Plot
with plt.style.context(["science", "nature"]):
    fig, ax = plt.subplots()
    ax.autoscale(tight=True)
    ax.set(**pparam)
    ax.set(title="Platt Scaling")
    ax.set(xlim=(f_hat.min(), f_hat.max()))
    ax.set(ylim=(-0.1, 1.1))
    ax.set(ylabel=r'$P(y=1|x)$')
    ax.scatter(f_hat, platt, c=g, cmap='bwr', s=1, alpha=0.5, label=r'$g(x)$')
    ax.axhline(0.5, color='purple', linestyle='--', alpha=0.5, label=r'$P(y=1|x)=0.5$')
    ax.legend()
    fig.savefig('platt.png', dpi=300, bbox_inches='tight')

# ROC Curve
with plt.style.context(["science", "nature"]):
    fig, ax = plt.subplots()
    ax.autoscale(tight=True)
    ax.set_aspect('equal')
    ax.set(**pparam)
    ax.set(title=f"ROC Curve (AUC = {auc:.3f})")
    ax.set(xlim=(-0.04, 1.04))
    ax.set(ylim=(-0.04, 1.04))
    ax.set(xlabel=r'FPR')
    ax.set(ylabel=r'TPR')
    ax.plot(fpr, tpr, color='r', label=r'Data')
    ax.plot([0, 1], [0, 1], color='b', linestyle=':', alpha=0.5, label=r'Random')
    ax.legend()
    fig.savefig('roc.png', dpi=300, bbox_inches='tight')