How to show the t-SNE？

[ZhihaoChe]

I am very excited to see such an excellent article. Could you please provide the code for the t-SNE graph?

[viyjy]

Hi, I probably used the following code.

import numpy as np
import matplotlib.pyplot as plt
from sklearn import datasets
from sklearn.manifold import TSNE
import matplotlib.cm as cm
import argparse
import os

class FeatureVisualize(object):
    '''
    Visualize features by TSNE
    '''
    def __init__(self, features, labels):
        '''
        features: (m,n)
        labels: (m,)
        '''
        self.features = features
        self.labels = labels

    def plot_tsne(self, save_path, save_eps=False):
        ''' Plot TSNE figure. Set save_eps=True if you want to save a .eps file.
        '''
        tsne = TSNE(n_components=2, init='pca', random_state=0)
        features = tsne.fit_transform(self.features)
        x_min, x_max = np.min(features, 0), np.max(features, 0)
        data = (features - x_min) / (x_max - x_min)
        del features
        
        colors = ["b", "r"]
        #plt.plot(data[:, 0], data[:, 1], color=self.labels 'bo')
        for i in range(data.shape[0]):
            color = colors[self.labels[i]]
            plt.scatter(data[i, 0], data[i, 1], color=color, s=10)

        plt.xticks([])
        plt.yticks([])
        if save_eps:
            plt.savefig(os.path.join(save_path, 'tsne_v1.png'), dpi=600, bbox_inches='tight')
        plt.show()


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--name", required=True,
                        help="Name of this run. Used for monitoring.")
    parser.add_argument("--dataset", default="svhn2mnist",
                        help="Which downstream task.")
    parser.add_argument("--save_dir", default="./tsne_plot", type=str,
                        help="The directory where attention maps will be saved.")
    args = parser.parse_args()

    save_path = os.path.join(args.save_dir, args.dataset, args.name)
    features = np.load(os.path.join(save_path, 'feature.npy'))
    labels = np.load(os.path.join(save_path, 'label.npy'))
    vis = FeatureVisualize(features, labels)
    vis.plot_tsne(save_path, save_eps=True)


if __name__ == '__main__':
    main()