Libsvm GUI

A simple graphical frontend for Libsvm mainly intended for didactic purposes. You can create data points by point and click and visualize the decision region induced by different kernels and parameter settings.
To create positive examples click the left mouse button; to create negative examples click the right button.
If all examples are from the same class, it uses a one-class SVM.
from __future__ import division, print_function
 
print(__doc__)
 
# Author: Peter Prettenhoer <peter.prettenhofer@gmail.com>
#
# License: BSD 3 clause
 
import matplotlib
matplotlib.use('TkAgg')
 
from matplotlib.backends.backend_tkagg import FigureCanvasTkAgg
from matplotlib.backends.backend_tkagg import NavigationToolbar2TkAgg
from matplotlib.figure import Figure
from matplotlib.contour import ContourSet
 
try:
    import tkinter as Tk
except ImportError:
    # Backward compat for Python 2
    import Tkinter as Tk
 
import sys
import numpy as np
 
from sklearn import svm
from sklearn.datasets import dump_svmlight_file
from sklearn.externals.six.moves import xrange
 
y_min, y_max = -50, 50
x_min, x_max = -50, 50
 
 
class Model(object):
    """The Model which hold the data. It implements the
    observable in the observer pattern and notifies the
    registered observers on change event.
    """
 
    def __init__(self):
        self.observers = []
        self.surface = None
        self.data = []
        self.cls = None
        self.surface_type = 0
 
    def changed(self, event):
        """Notify the observers. """
        for observer in self.observers:
            observer.update(event, self)
 
    def add_observer(self, observer):
        """Register an observer. """
        self.observers.append(observer)
 
    def set_surface(self, surface):
        self.surface = surface
 
    def dump_svmlight_file(self, file):
        data = np.array(self.data)
        X = data[:, 0:2]
        y = data[:, 2]
        dump_svmlight_file(X, y, file)
 
 
class Controller(object):
    def __init__(self, model):
        self.model = model
        self.kernel = Tk.IntVar()
        self.surface_type = Tk.IntVar()
        # Whether or not a model has been fitted
        self.fitted = False
 
    def fit(self):
        print("fit the model")
        train = np.array(self.model.data)
        X = train[:, 0:2]
        y = train[:, 2]
 
        C = float(self.complexity.get())
        gamma = float(self.gamma.get())
        coef0 = float(self.coef0.get())
        degree = int(self.degree.get())
        kernel_map = {0: "linear", 1: "rbf", 2: "poly"}
        if len(np.unique(y)) == 1:
            clf = svm.OneClassSVM(kernel=kernel_map[self.kernel.get()],
                                  gamma=gamma, coef0=coef0, degree=degree)
            clf.fit(X)
        else:
            clf = svm.SVC(kernel=kernel_map[self.kernel.get()], C=C,
                          gamma=gamma, coef0=coef0, degree=degree)
            clf.fit(X, y)
        if hasattr(clf, 'score'):
            print("Accuracy:", clf.score(X, y) * 100)
        X1, X2, Z = self.decision_surface(clf)
        self.model.clf = clf
        self.model.set_surface((X1, X2, Z))
        self.model.surface_type = self.surface_type.get()
        self.fitted = True
        self.model.changed("surface")
 
    def decision_surface(self, cls):
        delta = 1
        x = np.arange(x_min, x_max + delta, delta)
        y = np.arange(y_min, y_max + delta, delta)
        X1, X2 = np.meshgrid(x, y)
        Z = cls.decision_function(np.c_[X1.ravel(), X2.ravel()])
        Z = Z.reshape(X1.shape)
        return X1, X2, Z
 
    def clear_data(self):
        self.model.data = []
        self.fitted = False
        self.model.changed("clear")
 
    def add_example(self, x, y, label):
        self.model.data.append((x, y, label))
        self.model.changed("example_added")
 
        # update decision surface if already fitted.
        self.refit()
 
    def refit(self):
        """Refit the model if already fitted. """
        if self.fitted:
            self.fit()
 
 
class View(object):
    """Test docstring. """
    def __init__(self, root, controller):
        f = Figure()
        ax = f.add_subplot(111)
        ax.set_xticks([])
        ax.set_yticks([])
        ax.set_xlim((x_min, x_max))
        ax.set_ylim((y_min, y_max))
        canvas = FigureCanvasTkAgg(f, master=root)
        canvas.show()
        canvas.get_tk_widget().pack(side=Tk.TOP, fill=Tk.BOTH, expand=1)
        canvas._tkcanvas.pack(side=Tk.TOP, fill=Tk.BOTH, expand=1)
        canvas.mpl_connect('button_press_event', self.onclick)
        toolbar = NavigationToolbar2TkAgg(canvas, root)
        toolbar.update()
        self.controllbar = ControllBar(root, controller)
        self.f = f
        self.ax = ax
        self.canvas = canvas
        self.controller = controller
        self.contours = []
        self.c_labels = None
        self.plot_kernels()
 
    def plot_kernels(self):
        self.ax.text(-50, -60, "Linear: $u^T v$")
        self.ax.text(-20, -60, "RBF: $\exp (-\gamma \| u-v \|^2)$")
        self.ax.text(10, -60, "Poly: $(\gamma \, u^T v + r)^d$")
 
    def onclick(self, event):
        if event.xdata and event.ydata:
            if event.button == 1:
                self.controller.add_example(event.xdata, event.ydata, 1)
            elif event.button == 3:
                self.controller.add_example(event.xdata, event.ydata, -1)
 
    def update_example(self, model, idx):
        x, y, l = model.data[idx]
        if l == 1:
            color = 'w'
        elif l == -1:
            color = 'k'
        self.ax.plot([x], [y], "%so" % color, scalex=0.0, scaley=0.0)
 
    def update(self, event, model):
        if event == "examples_loaded":
            for i in xrange(len(model.data)):
                self.update_example(model, i)
 
        if event == "example_added":
            self.update_example(model, -1)
 
        if event == "clear":
            self.ax.clear()
            self.ax.set_xticks([])
            self.ax.set_yticks([])
            self.contours = []
            self.c_labels = None
            self.plot_kernels()
 
        if event == "surface":
            self.remove_surface()
            self.plot_support_vectors(model.clf.support_vectors_)
            self.plot_decision_surface(model.surface, model.surface_type)
 
        self.canvas.draw()
 
    def remove_surface(self):
        """Remove old decision surface."""
        if len(self.contours) > 0:
            for contour in self.contours:
                if isinstance(contour, ContourSet):
                    for lineset in contour.collections:
                        lineset.remove()
                else:
                    contour.remove()
            self.contours = []
 
    def plot_support_vectors(self, support_vectors):
        """Plot the support vectors by placing circles over the
        corresponding data points and adds the circle collection
        to the contours list."""
        cs = self.ax.scatter(support_vectors[:, 0], support_vectors[:, 1],
                             s=80, edgecolors="k", facecolors="none")
        self.contours.append(cs)
 
    def plot_decision_surface(self, surface, type):
        X1, X2, Z = surface
        if type == 0:
            levels = [-1.0, 0.0, 1.0]
            linestyles = ['dashed', 'solid', 'dashed']
            colors = 'k'
            self.contours.append(self.ax.contour(X1, X2, Z, levels,
                                                 colors=colors,
                                                 linestyles=linestyles))
        elif type == 1:
            self.contours.append(self.ax.contourf(X1, X2, Z, 10,
                                                  cmap=matplotlib.cm.bone,
                                                  origin='lower', alpha=0.85))
            self.contours.append(self.ax.contour(X1, X2, Z, [0.0], colors='k',
                                                 linestyles=['solid']))
        else:
            raise ValueError("surface type unknown")
 
 
class ControllBar(object):
    def __init__(self, root, controller):
        fm = Tk.Frame(root)
        kernel_group = Tk.Frame(fm)
        Tk.Radiobutton(kernel_group, text="Linear", variable=controller.kernel,
                       value=0, command=controller.refit).pack(anchor=Tk.W)
        Tk.Radiobutton(kernel_group, text="RBF", variable=controller.kernel,
                       value=1, command=controller.refit).pack(anchor=Tk.W)
        Tk.Radiobutton(kernel_group, text="Poly", variable=controller.kernel,
                       value=2, command=controller.refit).pack(anchor=Tk.W)
        kernel_group.pack(side=Tk.LEFT)
 
        valbox = Tk.Frame(fm)
        controller.complexity = Tk.StringVar()
        controller.complexity.set("1.0")
        c = Tk.Frame(valbox)
        Tk.Label(c, text="C:", anchor="e", width=7).pack(side=Tk.LEFT)
        Tk.Entry(c, width=6, textvariable=controller.complexity).pack(
            side=Tk.LEFT)
        c.pack()
 
        controller.gamma = Tk.StringVar()
        controller.gamma.set("0.01")
        g = Tk.Frame(valbox)
        Tk.Label(g, text="gamma:", anchor="e", width=7).pack(side=Tk.LEFT)
        Tk.Entry(g, width=6, textvariable=controller.gamma).pack(side=Tk.LEFT)
        g.pack()
 
        controller.degree = Tk.StringVar()
        controller.degree.set("3")
        d = Tk.Frame(valbox)
        Tk.Label(d, text="degree:", anchor="e", width=7).pack(side=Tk.LEFT)
        Tk.Entry(d, width=6, textvariable=controller.degree).pack(side=Tk.LEFT)
        d.pack()
 
        controller.coef0 = Tk.StringVar()
        controller.coef0.set("0")
        r = Tk.Frame(valbox)
        Tk.Label(r, text="coef0:", anchor="e", width=7).pack(side=Tk.LEFT)
        Tk.Entry(r, width=6, textvariable=controller.coef0).pack(side=Tk.LEFT)
        r.pack()
        valbox.pack(side=Tk.LEFT)
 
        cmap_group = Tk.Frame(fm)
        Tk.Radiobutton(cmap_group, text="Hyperplanes",
                       variable=controller.surface_type, value=0,
                       command=controller.refit).pack(anchor=Tk.W)
        Tk.Radiobutton(cmap_group, text="Surface",
                       variable=controller.surface_type, value=1,
                       command=controller.refit).pack(anchor=Tk.W)
 
        cmap_group.pack(side=Tk.LEFT)
 
        train_button = Tk.Button(fm, text='Fit', width=5,
                                 command=controller.fit)
        train_button.pack()
        fm.pack(side=Tk.LEFT)
        Tk.Button(fm, text='Clear', width=5,
                  command=controller.clear_data).pack(side=Tk.LEFT)
 
 
def get_parser():
    from optparse import OptionParser
    op = OptionParser()
    op.add_option("--output",
                  action="store", type="str", dest="output",
                  help="Path where to dump data.")
    return op
 
 
def main(argv):
    op = get_parser()
    opts, args = op.parse_args(argv[1:])
    root = Tk.Tk()
    model = Model()
    controller = Controller(model)
    root.wm_title("Scikit-learn Libsvm GUI")
    view = View(root, controller)
    model.add_observer(view)
    Tk.mainloop()
 
    if opts.output:
        model.dump_svmlight_file(opts.output)
 
if __name__ == "__main__":
    main(sys.argv)
Total running time of the script: (0 minutes 0.000 seconds)
Download Python source code: svm_gui.py
Download IPython notebook: svm_gui.ipynb
Links:
http://scikit-learn.org/stable/auto_examples/applications/svm_gui.html
doc_scikit_learn
2025-01-10 15:47:30
Comments