Thomas Baruchel baruchel

Implementing APL rules for broadcasting into Numpy

This is an attempt for extending broadcasting rules in Numpy without losing the benefits of vectorization by converting new rules to preexisting ones.

The broadcasting policy is taken from APL standards according to the following ideas:

a subset of nested arrays from APL is implemented here;
an axis parameter is implemented here.

Enclosing parts of an array

I previously wrote a plugin called vim-notebook for using Maxima in a Vim buffer. But since Maxima is written in Lisp and intended to be usable from Lisp code, I finally figured out a very nice way to achieve a similar result by using the existing Slimv.

In this post, I explain how to:

install a Lisp implementation (Clozure CL);
install the Quicklisp library manager;
install the Swank server;
fetch the latest version of Maxima and compile the interpreter only;
integrate Maxima to Vim with Slimv.

Each time I re-install a fresh Linux distribution, one of the first things I need to do is to compile the Heirloom Documentation Tools. However some adjustments have to be performed; I write them here in order to recall them later:

download the sources from http://heirloom.sourceforge.net/doctools.html
type: sudo apt-get install g++ libc-dev bison flex
uncompress the tarball
enter the heirloom-doctools directory
edit mk.config and change three lines:
- replace INSTALL=/usr/ucb/install with INSTALL=/usr/bin/install

replace PREFIX=/usr with PREFIX=/opt/heirloom

Hacking Dyalog APL # 1

Dyalog APL is a great tool but the interface for working in an interactive session is rather poor. Editing previous lines by moving the cursor to some other location is rather painful. Of course Dyalog APL isn't allowed to use the great readline library. But it is not too difficult to replace the officiel ncurses interface by some more modern readline-based one as long as you do it with an external tool.

What you will get is:

a modern readline-based input;
a great history system (persistent accross sessions if you wish);
the ability to launch your favorite editor on the fly for editing a more complex line of input (or defining a traditional function) whenever you want;
leaving Dyalog APL with the Ctrl-D signal;

Thomas Baruchel's mathematical notebook

(https://gist.github.com/2b870728323456f71402)

This is my mathematical notebook; I put here the identities I find from time to time; this notebook is written in the markdown format with the vim editor. I use my own vim-notebook plugin for embedding Maxima code in it; thus I can evaluate each cell of code.

	# Adapted from:
	# "A Floating-Point Technique for Extending the vailable Precision"
	# by T.J. Dekker in Numer. Math. 18, 224-242, 1971.
	# (adapted to Numpy from the Algol 60 code in the paper)

	import numpy as np

	import decimal
	_npDecimal = np.vectorize(decimal.Decimal, otypes=[object])

	# -- coding: utf8 --

	def library_sort(l):
	# Initialization
	d = len(l)
	k = [None]*(d<<1)
	m = d.bit_length() # floor(log2(n) + 1)
	for i in range(d): k[2*i+1] = l[i]

	# main loop

	# -- coding: utf-8 --

	import numpy as np
	import scipy.ndimage.measurements
	import random

	def kMeans( data, k, centers=None, iter=64 ):
	if len(data.shape) == 1:
	data = np.vstack(data)

	" quickmove.vim : move the cursor according to the more/less game
	" Author: Thomas Baruchel
	" Version: 1.0
	" Date: Mar 14, 2015
	"

	let g:quickmove_version = "1.0"
	if &compatible
	finish
	endif

	-- Reproduire la table `CountryLanguage` à la différence près que la colonne Percentage doit maintenant contenir le pourcentage de locuteurs d'une langue dans un pays par rapport au nombre total de locuteurs de cette langue dans le monde.

	SELECT CountryCode, Language, IsOfficial,
	(A.Percentage * A.Population / B.nbr) AS Percentage
	FROM (
	SELECT CountryLanguage.*, Country.Population
	FROM CountryLanguage
	LEFT JOIN Country
	ON CountryLanguage.CountryCode = Country.Code
	) AS A