def embedded_from_tsne(self, model_filename, num_components=2): self._parse_model_parameters(model_filename) params = autoenc_params.setupModelFromFlags() sess = params[0] autoenc = params[3] model_filename = params[7] output, encoding = sess.run([autoenc.getDecoding(True), autoenc.encoding_layers[-1]], feed_dict={autoenc.input_placeholder : self.real_images}) self.embedded_images = TSNE(n_components=num_components, perplexity=40, verbose=2).fit_transform(encoding)
def encodings_from_model(self, model_filename): self._parse_model_parameters(model_filename) params = autoenc_params.setupModelFromFlags() sess = params[0] autoenc = params[3] model_filename = params[7] output, encoding = sess.run([autoenc.getDecoding(True), autoenc.encoding_layers[-1]], feed_dict={autoenc.input_placeholder : self.real_images}) self.encoded_images = encoding
FLAGS = tf.app.flags.FLAGS image_size = 20 if FLAGS.dataset.lower() == "smiley" else 28 app = QtGui.QApplication([]) ## Create window with ImageView widget win = QtGui.QWidget() layout = QtGui.QGridLayout() win.setLayout(layout) win.resize(800,800) imv = pg.ImageView() print win.layout() win.layout().addWidget(imv, 1, 0) sess, global_step, train_step, autoenc, loss, writer, saver, model_filename, cheap_summaries, expensive_summaries, test_loss_summary, classifier_out, y = autoenc_params.setupModelFromFlags() # Add sliders to manipulate the encoding def displayImage(encodedActivations): print autoenc.encoding_layers[-1].get_shape() print autoenc.encoding_layers[-1] enc = np.matrix(encodedActivations) output = sess.run(autoenc.getDecoding(True), feed_dict={autoenc.encoding_layers[-1] : enc}) imv.setImage(output.reshape(image_size, image_size).T) sliders = [] def sliderModev(val): encAct = [sl.value()/10. for sl in sliders] print encAct displayImage(encAct)