rmmh

Crosspoint Reader Dictionary Index Experiments

The .idx file (StarDict/dictd format) is just a flat, alphabetically-sorted list of variable-length entries — null-terminated word, then a 4-byte big-endian offset and 4-byte size into the .dict file. Because entries aren't fixed-width, you can't binary-search the raw file directly (you can't jump to "entry N" without scanning from the start, or doing random reads and heuristics), so a naive lookup means either loading the whole index into RAM or linear-scanning it block by block, which is slow on an SD card over an SPI interface.

The general goal is to minimize the number of 512B sectors read from the SD card to find a word, since SD card reads are significantly slower than RAM reads. We need to get as much value from each SD read as possible.

This benchmark builds and compares several small sidecar index structures (oft, cspt, base26, fence/fencec, fencep, tab) that let you binary-search a compact auxiliary table instead, narrowing down to a small byte ran

rmmh

#!/usr/bin/env python3

import os
import time
import traceback

import pychromecast

import RPi.GPIO as GPIO
from mfrc522 import MFRC522

rmmh

#!/usr/bin/env python3

# grab data from https://place.thatguyalex.com/

import glob

import numpy
import PIL.Image

import scipy.stats

rmmh

function dec(xs) {
    function n2w(n) {
        var o = '';
        for (var i = 0; i < 5; i++) {
            var c = n % 26;
            n = (n / 26)|0;
            o += 'MGXJQZVBDCTWHFKSPNLREOAIUY'[c];
        }
        return o;
    }

rmmh

https://www.facebook.com/watch/live/?ref=watch_permalink&v=1264023127446649

Welcome to Connect 2021.

My process for these talks is that I basically go back through last year of internal posts and public tweets that I’ve made looking for things that are interesting and worthwhile to talk about ask then I try to roughly sort them in to some kind of a coherent topic flow. I’m also usually the one that’s lot more kind of grumpy and unhappy with the way things are going as I always want more and faster progress. But this year, I really have a lot of things to be quite happy about. By far the most important is that quest 2 has been a really big success and it was a heroic effort to get it out when it was last year in the space of all the headwinds that we had, but it was better, faster, cheaper, one of those just rare combinations that you almost never get to have in a product.

00:01:31 While there’s still a couple minor things where some people miss the displays or continuous IPD adjustments, it was a fantastic

rmmh

#!/usr/bin/env python3

import re
import urllib

import dateutil
import requests_cache

session = requests_cache.CachedSession(allowable_codes=(200, 404))
session.headers['User-Agent'] = 'CovidDeathsPopularity/1.0 (hitchmanr@gmail.com) python3/requests'

rmmh

#!/usr/bin/env python
# Copyright 2016 The Kubernetes Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software

rmmh

// How many distinct closed paths visit each cell in an NxN grid exactly once?

package main

import (
	"fmt"
	"math/bits"
)

type walls struct {

rmmh

#!/usr/bin/env python3

# an experiment in using a large corpus of phrases to generate initialisms from

import re
import zipfile

from nltk import tokenize

phrases = {}

rmmh

import org.lwjgl.opengl.GL11;

public abstract class bb {

    protected pd b;
    private ic a = new dx();
    private bw e = new bw();
    protected float c = 0.0F;
    protected float d = 1.0F;