sklearn-docs/SVM-Kernels
3
1# Code source: Gaël Varoquaux2# License: BSD 3 clause3 4import numpy as np5import matplotlib.pyplot as plt6from sklearn import svm7import gradio as gr8from matplotlib.colors import ListedColormap9plt.switch_backend("agg")10 11font1 = {'family':'DejaVu Sans','size':20}12 13def create_data(random, size_num, x_min, x_max, y_min, y_max):14 #emulate some random data15 if random:16 size_num = int(size_num)17 x = np.random.uniform(x_min, x_max, size=(size_num, 1))18 y = np.random.uniform(y_min, y_max, size=(size_num, 1))19 20 X = np.hstack((x, y))21 Y = [0] * int(size_num/2) + [1] * int(size_num/2)22 else:23 X = np.c_[24 (0.4, -0.7),25 (-1.5, -1),26 (-1.4, -0.9),27 (-1.3, -1.2),28 (-1.5, 0.2),29 (-1.2, -0.4),30 (-0.5, 1.2),31 (-1.5, 2.1),32 (1, 1),33 # --34 (1.3, 0.8),35 (1.5, 0.5),36 (0.2, -2),37 (0.5, -2.4),38 (0.2, -2.3),39 (0, -2.7),40 (1.3, 2.8),41 ].T42 43 Y = [0] * 8 + [1] * 844 return X, Y45 46# fit the model47def clf_kernel(color1, color2, dpi, size_num = None, x_min = None, 48 x_max = None, y_min = None,49 y_max = None, random = False):50 51 if size_num is not None or x_min is not None or x_max is not None or y_min is not None or y_max is not None:52 random = True53 54 X, Y = create_data(random, size_num, x_min, x_max, y_min, y_max)55 56 kernels = ["linear", "poly", "rbf"]57 58 # plot the line, the points, and the nearest vectors to the plane 59 fig, axs = plt.subplots(1,3, figsize = (16,8), facecolor='none', dpi = res[dpi])60 61 cmap = ListedColormap([color1, color2], N=2, name = 'braincell')62 for i, kernel in enumerate(kernels):63 clf = svm.SVC(kernel=kernel, gamma=2)64 clf.fit(X, Y)65 axs[i].scatter(66 clf.support_vectors_[:, 0],67 clf.support_vectors_[:, 1],68 s=80,69 facecolors="none",70 zorder=10,71 edgecolors="k",72 )73 axs[i].scatter(X[:, 0], X[:, 1], c=Y, zorder=10, cmap=cmap, edgecolors="k")74 75 axs[i].axis("tight")76 x_min = -377 x_max = 378 y_min = -379 y_max = 380 81 XX, YY = np.mgrid[x_min:x_max:200j, y_min:y_max:200j]82 Z = clf.decision_function(np.c_[XX.ravel(), YY.ravel()])83 84 # Put the result into a color plot85 Z = Z.reshape(XX.shape)86 axs[i].pcolormesh(XX, YY, Z > 0, cmap=cmap)87 axs[i].contour(88 XX,89 YY,90 Z,91 colors=["k", "k", "k"],92 linestyles=["--", "-", "--"],93 levels=[-0.5, 0, 0.5],94 )95 96 axs[i].set_xlim(x_min, x_max)97 axs[i].set_ylim(y_min, y_max)98 99 axs[i].set_xticks(())100 axs[i].set_yticks(())101 axs[i].set_title('Type of kernel: ' + kernel, 102 color = "white", fontdict = font1, pad=20, 103 bbox=dict(boxstyle="round,pad=0.3", 104 color = "#6366F1"))105 106 plt.close()107 return fig, np.round(X, decimals=2)108 109intro = """<h1 style="text-align: center;">🤗 Introducing SVM-Kernels 🤗</h1>110"""111desc = """<h3 style="text-align: center;">Three different types of SVM-Kernels are displayed below. 112The polynomial and RBF are especially useful when the data-points are not linearly separable. </h3>113"""114notice = """<br><div style = "text-align: left;"> <em>Notice: Run the model on example data or use <strong>Randomize data</strong> 115button below to check out the model on randomized data-points. Any changes to visual parameters will reset the data!</em></div>"""116 117notice2 = """<br><div style = "text-align: left;"> <em>Notice: The data points are categorized into two distinct classes, and they are evenly distributed on the plots to visually represent these classes.</em></div>"""118 119made ="""<div style="text-align: center;">120 <p>Made with ❤</p>"""121 122link = """<div style="text-align: center;">123<a href="https://scikit-learn.org/stable/auto_examples/svm/plot_svm_kernels.html#sphx-glr-auto-examples-svm-plot-svm-kernels-py" target="_blank" rel="noopener noreferrer">124Demo is based on this script from scikit-learn documentation</a>"""125 126res = {'Small': 50, 'Medium': 75, 'Large': 100}127 128with gr.Blocks(theme=gr.themes.Soft(primary_hue="indigo",129 secondary_hue="violet",130 neutral_hue="slate",131 font = gr.themes.GoogleFont("Inter")),132 title="SVM-Kernels") as demo:133 134 gr.HTML(intro)135 gr.HTML(desc)136 137 with gr.Tab("Plotted results"):138 plot = gr.Plot(label="Kernel comparison:")139 with gr.Tab("Data coordinates"):140 gr.HTML(notice2)141 X = gr.Numpy(headers = ['x','y'], interactive=False)142 143 with gr.Column():144 145 with gr.Accordion(label = 'Randomize data'):146 gr.HTML(notice)147 samples = gr.Slider(4, 16, value = 8, step = 2, label = "Number of samples:")148 x_min = gr.Slider(-3, 0, value=-2, step=0.1, label="X Min:")149 x_max = gr.Slider(0, 3, value=2, step=0.1, label="X Max:")150 y_min = gr.Slider(-3, 0, value=-2, step=0.1, label="Y Min:")151 y_max = gr.Slider(0, 3, value=2, step=0.1, label="Y Max:")152 random = gr.Button("Randomize data")153 154 155 156 157 with gr.Accordion(label = "Visual parameters"):158 with gr.Row():159 color1 = gr.ColorPicker(label = 'Pick color one:', value = '#9abfd8')160 color2 = gr.ColorPicker(label = 'Pick color two:', value = '#371c4b')161 #dpi = gr.Slider(50, 100, value = 75, step = 1, label = "Set the resolution: ")162 dpi = gr.Radio(list(res.keys()), value = 'Medium', label = "Select the plot size:")163 164 params2 = [color1, color2, dpi]165 166 random.click(fn=clf_kernel, inputs=[color1, color2, dpi,samples, x_min, x_max, y_min, y_max], outputs=[plot,X]) 167 168 for i in params2:169 i.change(fn=clf_kernel, inputs=[color1, color2,dpi], outputs=[plot, X])170 171 demo.load(fn=clf_kernel, inputs=[color1, color2, dpi], outputs=[plot,X]) 172 gr.HTML(made)173 gr.HTML(link)174 175demo.launch()