markloyman’s gists

markloyman / hubness.py

Created May 2, 2020 11:13

Hubness calculation (index defined by Miloš Radovanovi´c et al, 2010)

	import numpy as np
	from scipy.stats import skew


	def k_occurrences(nbrs_indices):
	"""
	Calculate k-occurrences distribution
	Miloš Radovanovi´c et al,
	Hubs in space: Popular nearest neighbors in high-dimensional data.
	Journal of Machine Learning Research, 11(Sep):2487–2531, 2010

markloyman / pearson_loss.py

Created May 2, 2020 10:15

Keras Pearson Correlation Loss

	def l2_distance_matrix(x, exact=True):
	"""
	Calculate an L2 distance matrix from embedding vectors)
	:param
	x: embedding tensor size (batch_size, embedding_length)
	exact: if False, skip the final sqrt
	:return: l2 distance matrix tensor tensor size (batch_size, batch_size)
	"""
	r = K.sum(x*x, 1)
	r = K.reshape(r, [-1, 1]) # turn r into column vector

markloyman / cost_volume_viz.py

Last active October 23, 2018 06:58

Cost Volume Visualization (for stereo depth estimation debuging and exploration)

	import numpy as np
	import matplotlib.pyplot as plt
	from volume_viz import VolumeViz


	class CostVolumeViz:
	def __init__(self, left, right, disparity):
	self.left = left
	self.right = right
	self.disparity = self.StandardizeDisparity(disparity)

markloyman / volume_viz.py

Created September 21, 2018 09:04

prototype for interactive 3d volume visualization

	import numpy as np
	import matplotlib.pyplot as plt

	class VolumeViz:
	def __init__(self, fig, volume):
	self.volume = volume
	self.figure = fig
	self.ax1 = fig.add_subplot(121)
	self.ax2 = fig.add_subplot(122)
	self.ax1.set_title('click')

markloyman / runtime_stats_keras.py

Created September 5, 2018 12:17

Use TensorBoard's runtime statistcs with a Keras model

	import numpy as np
	import tensorflow as tf
	sess = tf.Session()
	from keras import backend as K
	K.set_session(sess)
	from keras.objectives import mean_squared_error
	from keras.layers import Dense


	def load_dummy_data(n = 1000):

markloyman / triplet_loss.py

Created January 23, 2018 20:50

Triplet Loss

	def triplet_loss(_, y_pred, triplet_margin=1):
	'''
	Assume: y_pred shape is (batch_size, 2)

	Example for how to construct such a corresponding triplet network:

	def euclidean_distance(vects):
	x, y = vects
	return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon()))

markloyman / crop_random_with_mask.py

Created September 28, 2017 19:58

Random crop of an image and it's mask (crop is aware of mask region)

	def crop(image, mask, min_size= 128):
	# Determine New Size on Y axis
	max_size_y = image.shape[0]
	mask_y = np.where(np.sum(mask, axis=1))
	min_size_y = np.max(mask_y) - np.min(mask_y)
	new_size_y = np.random.rand() * (max_size_y - min_size_y)
	new_size_y = np.maximum(new_size_y, min_size).astype('uint16')
	# Determine New Size on X axis
	max_size_x = image.shape[1]
	mask_x = np.where(np.sum(mask, axis=0))

Mark Loyman markloyman