This package was designed to present classification boundary methods. The three methods include broad initial sampling, low point-density sampling, and path tracing. The following code demonstrates how to use each module to run classification boundary simulations. Here we use a classification polygon to determine the classification boundary.
(require 'ox-ipynb)This will import the boundary search package and all the sub-modules within it. It will also import matplotlib’s Polygon and Path for use in the classification tasks.
import os
os.system("pip install git+https://github.com/mayabhat/boundary_search.git --upgrade")
from sbs.densitysampling import sample as dsam
from sbs.pathtracing import sample as psam
from sbs.broadsampling import sample as bsam
from shapely.geometry import Polygon
import matplotlib.path as mplPath
import matplotlib.pyplot as plt
import numpy as np
from sklearn import svm
polygon = Polygon([(1, 1),(1, 2),(2, 2),(2, 1)])
x,y = polygon.exterior.xy
fig, ax = plt.subplots(figsize=(5,5))
ax.plot(x, y)
ax.set_xlim((0, 3))
ax.set_ylim((0, 3))
ax.set_ylabel('x2')
ax.set_xlabel('x1')
plt.show()
bs = bsam([[1.5, 1.5]], polygon, boundmin=0, boundmax=3)
bs.first_sample(ni=16)
bs.iter_sample(min_points=2, samples=4)
bs.plot_final(bs.X, bs.cat)
To initialize, bsam takes a center point, a polygon shape, and a minimum and maximum bound. The first sample generates a distribution of points around the center point provided. ni is a hyperparameter that determines how many points to initially sample. Other parameters can be found in the broad initial sampling py file in sbs.
iter_sample performs the simulation to recover the classification boundary within a convergence criteria. The convergence criteria are also hyperparameters that can be specified.
plot_final plots the true boundary, sample points and their respective classes, predicted boundary, and 95% confidence interval.
See documentation for hyperparameters.
ds = dsam([1, 1], polygon, 0, 3)
ds.first_sample(dis=0.2)
ds.iter_sample(tol=0.01, n=4, rad_den=0.2, conv_trials=4,
min_points=1, domain_step=0.15)
ds.plot_final(ds.X, ds.cat)
To initialize, dsam takes a center point, a polygon shape, and a minimum and maximum bound. The first sample generates a distribution of points around the center point provided. dis is a hyperparameter that determines the initial radius for sample point distribution. Other parameters can be found in the broad initial sampling py file in sbs.
iter_sample performs the simulation to recover the classification boundary within a convergence criteria. The convergence criteria are also hyperparameters that can be specified.
plot_final plots the true boundary, sample points and their respective classes, predicted boundary, and 95% confidence interval.
See documentation for hyperparameters.
ps = psam([[1, 1]], polygon, 0, 3)
ps.first_sample(dis=0.3, seed=10)
ps.iter_sample(step=ps.scale*2, dis=0.1, ran_sam=False,
conv_num=3, centol = ps.scale*3, seed=10)
ps.plot_final(ps.X, ps.cat)
To initialize, psam takes a center point, a polygon shape, and a minimum and maximum bound. The first sample generates a distribution of points around the center point provided. dis is a hyperparameter that determines the initial radius for sample point distribution. Other parameters can be found in the broad initial sampling py file in sbs.
iter_sample performs the simulation to recover the classification boundary within a convergence criteria. The convergence criteria are also hyperparameters that can be specified.
plot_final plots the true boundary, sample points and their respective classes, predicted boundary, and 95% confidence interval.
See documentation for hyperparameters.