Module src.app.Clustering.DimReductionTSNE

Source code
from src.app.Module import Module
from sklearn.manifold import TSNE
import cv2
import numpy as np
import matplotlib.pyplot as plt


class DimReductionTSNE(Module):
    """Applies t-SNE dimensionality reduction to higher dimensional clustering results.
    """
    def __init__(self, prev_module):
        super().__init__('DimReductionTSNE', prev_module)

    def run(self):
        super().run()
        features = self._data['features']
        centers = self._data['centers']
        features = np.concatenate((features, centers))

        tsne = TSNE(n_components=2, verbose=1, perplexity=40, n_iter=300)
        tsne_result = tsne.fit_transform(features)

        # Normalize results
        tsne_result[:, 0] = (-np.min(tsne_result[:, 0]) + tsne_result[:, 0]) / (-np.min(tsne_result[:, 0]) + np.max(tsne_result[:, 0]))
        tsne_result[:, 1] = (-np.min(tsne_result[:, 1]) + tsne_result[:, 1]) / (-np.min(tsne_result[:, 1]) + np.max(tsne_result[:, 1]))

        # Extract coordinates
        tsne1 = tsne_result[:len(self._data['features']), 0]
        tsne2 = tsne_result[:len(self._data['features']), 1]
        tsnec1 = tsne_result[len(self._data['features']):, 0]
        tsnec2 = tsne_result[len(self._data['features']):, 1]

        self._result = {
            'images': self._data['images'],
            'labels': self._data['labels'],
            'cx': tsnec1,
            'cy': tsnec2,
            'x': tsne1,
            'y': tsne2
        }

    def visualize(self, glyph_size=1):
        result = self.get_module_results()
        plt.figure()
        plt.axis([np.min(result['x']), np.max([result['x']]), np.min(result['y']), np.max(result['y'])])
        plt.scatter(result['x'], result['y'], c=result['labels'], s=0.5, cmap='Dark2')
        for i, image in enumerate(result['images']):
            img = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            extent = [result['x'][i]-(glyph_size/2), result['x'][i]+(glyph_size/2), result['y'][i]-(glyph_size/2), result['y'][i]+(glyph_size/2)]
            plt.imshow(img, origin='upper', extent=extent, cmap='gray')
        plt.savefig('graph.pdf', dpi=800)
        plt.show()

Classes

class DimReductionTSNE (prev_module)

Applies t-SNE dimensionality reduction to higher dimensional clustering results.

Source code
class DimReductionTSNE(Module):
    """Applies t-SNE dimensionality reduction to higher dimensional clustering results.
    """
    def __init__(self, prev_module):
        super().__init__('DimReductionTSNE', prev_module)

    def run(self):
        super().run()
        features = self._data['features']
        centers = self._data['centers']
        features = np.concatenate((features, centers))

        tsne = TSNE(n_components=2, verbose=1, perplexity=40, n_iter=300)
        tsne_result = tsne.fit_transform(features)

        # Normalize results
        tsne_result[:, 0] = (-np.min(tsne_result[:, 0]) + tsne_result[:, 0]) / (-np.min(tsne_result[:, 0]) + np.max(tsne_result[:, 0]))
        tsne_result[:, 1] = (-np.min(tsne_result[:, 1]) + tsne_result[:, 1]) / (-np.min(tsne_result[:, 1]) + np.max(tsne_result[:, 1]))

        # Extract coordinates
        tsne1 = tsne_result[:len(self._data['features']), 0]
        tsne2 = tsne_result[:len(self._data['features']), 1]
        tsnec1 = tsne_result[len(self._data['features']):, 0]
        tsnec2 = tsne_result[len(self._data['features']):, 1]

        self._result = {
            'images': self._data['images'],
            'labels': self._data['labels'],
            'cx': tsnec1,
            'cy': tsnec2,
            'x': tsne1,
            'y': tsne2
        }

    def visualize(self, glyph_size=1):
        result = self.get_module_results()
        plt.figure()
        plt.axis([np.min(result['x']), np.max([result['x']]), np.min(result['y']), np.max(result['y'])])
        plt.scatter(result['x'], result['y'], c=result['labels'], s=0.5, cmap='Dark2')
        for i, image in enumerate(result['images']):
            img = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            extent = [result['x'][i]-(glyph_size/2), result['x'][i]+(glyph_size/2), result['y'][i]-(glyph_size/2), result['y'][i]+(glyph_size/2)]
            plt.imshow(img, origin='upper', extent=extent, cmap='gray')
        plt.savefig('graph.pdf', dpi=800)
        plt.show()

Ancestors

Inherited members