radarsat1

# Steve's adaptation of VEEGAN example:
# https://github.com/akashgit/VEEGAN/blob/master/VEEGAN_2D_RING.ipynb
# to TensorFlow 2 + Keras, with the TensorFlow Probabilities library:
# https://blog.tensorflow.org/2019/03/variational-autoencoders-with.html
#
# Stephen Sinclair <radarsat1@gmail.com>

# Google Colab version here:
# https://colab.research.google.com/drive/1dSW7Yn8okef-ng-Inkev205L2Gy6_MIx

radarsat1

/* Test for collision points locations generated by purterbation rotations in Bullet Physics */

#define BT_USE_DOUBLE_PRECISION

#include <BulletCollision/CollisionDispatch/btCollisionObject.h>
#include <BulletCollision/CollisionDispatch/btCollisionWorld.h>
#include <BulletCollision/CollisionDispatch/btCollisionDispatcher.h>
#include <BulletCollision/CollisionDispatch/btDefaultCollisionConfiguration.h>
#include <BulletCollision/CollisionShapes/btBoxShape.h>
#include <BulletCollision/CollisionShapes/btConvexHullShape.h>

radarsat1

# A Python translation of the Brownian Surface function from
# http://www.mathworks.com/matlabcentral/fileexchange/38945-fractional-brownian-field-or-surface-generator
#
# More info, check https://en.wikipedia.org/wiki/Brownian_surface
#
# Example output: http://i.imgur.com/MKHxA6N.png

import numpy as np

#%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

radarsat1

#!/bin/bash
# tile four videos using ffmpeg

VID1="$1"
VID2="$2"
VID3="$3"
VID4="$4"
OUT="$5"

# ffmpeg's complex filter graph is powerful!

radarsat1

#!/usr/bin/env python

"""Provide a parallelized version of scipy.optimize.brute for 1-, 2-, or 3-dimensional arguments.

Needed to parallelize the steps of a grid-based global optimization, so I copied the "brute" code, replaced "vectorize" with nested maps for specific numbers of arguments, and replaced the outer-most map() with a ThreadPool.map().  Likely this could be generalized to any number of arguments (e.g. by zipping the grid and iterating in parallel over the tuples) but I didn't feel like figuring that out just now.

This should perform help for routines that spend a lot of time in numpy code, but I imagine that a pure Python routine might get caught-up by the GIL.  Possibly using a process pool instead of a thread pool would fix it.  Currently I'm using it for a routine that spends all its time in fftconvolve() so it wasn't a problem for me.

parbrute() takes the same arguments as scipy.optimize.brute(), with the addition of thread=N for the number of threads you want in the pool.  Defaults to 4