Howdy, I think this might appeal to some of you, sorry if it comes off a bit spammy. I've started a Discord server mostly for discussion of books (Other things too...our rule is if it's text-based or if it's printed on a page, it's allowed. We welcome interactive fiction, comics, poetry, theory, visual novels, etc.) that is aiming to keep discussion at least somewhat serious and a respectful atmosphere similar to what Tildes aims for. Eventually we'll do wider recruitment (posters in universities, probably), but for now we're trying to get a decent server culture going with people we can trust to not shit all over everything. If this sounds appealing, we'd love to have you :)

Paste with rules, should look familiar.

Permanent link for those who would like to join: https://discord.gg/yr4pA96

Here's the ruby gem page and here's the github. Right now it comes with a command line browser that can browse the front page and group pages with no sorting options, and you can view the contents of a topic (link or text) aswell as the comments. The methods defined in lib/tilde-scraper/api.rb can be used to scrape tildes pages into Group, Page, Topic, and Comment objects.

Right now it's super basic and messy, but I figured if anyone was interested in it it would be the people here.

https://gitlab.com/smoores/pliant

I’ve been a software developer for about three years, and I’ve always been enticed by and passionate about the open source scene. I have an assortment of projects variously available on GitHub and GitLab, but this is the first time I’ve ever created an open source project intended to be used by others.

Pliant is a barebones starter kit for anyone wanting to self host their own blog. It came out of my own efforts to start a blog, and it’s what currently powers https://tfhe.shanemoore.me.

I’d love to hear you’re feedback, or just discuss open source, blogging, web technologies, or whatever else comes up.

Hi there, I've started a new public server on Discord for music discussion, recommendations, etc. I've seen a number of these go down in flames or completely lose the appeal for their core audience, so I've borrowed some principles from Tildes, adapted for the somewhat different but related issues this kind of Discord server typically runs into. I thought this may be of interest to some users here, so you're all welcome as long as you play by our rules!

If you'd like to see our rules and goals first, check the paste here. They'll be pretty familiar to anyone here, though they are subject to change depending on our needs. This should go without saying, but to be clear, this server is not officially connected to Tildes, nor is it exclusive to Tildes users. There just might be some overlap in ideas :)

Here's our permanent invite link if you'd like to join: https://discord.gg/kC4sSQq

I am working on an app called Percent Done that is a combination of goal setting, time tracking and habit tracking.

I like setting time-based goals for myself every day, such as “write for an hour” or “work on Percent Done for four hours.”

I also like Seinfeld’s “don’t break the chain” method. For example, Apple Watch shows you how many days you have completed your exercise circle and tells you that you have been keeping at it for x days.

Percent Done is a marriage of these two concepts. It allows you to set goals and track the time you spend on them, as well as how many days in a row you have consistently completed them. For example, you can add a goal that says “write for an hour every day,” and Percent Done will notify you every day to write for an hour. You will be able to tap on this goal and Percent Done will start counting back from one hour. You will also be able to see how many days in a row you have written for an hour.

Here is a demo GIF.

You can also add one-time goals to Percent Done with or without time tracking, so it is a task management tool as well.

You can play with the design prototype here: Percent Done design prototype

I would really love to get your feedback on this. If you are interested in being a beta tester, feel free to reply to this topic or e-mail me at "hi at evrim dot io."

By the way, this is almost completely a self-promotion post. If it is against the rules, I'd be happy to remove this.

Introduction to Genetic Algorithms

Genetic algorithms can be used to solve problems that are difficult, or impossible to solve with traditional algorithms. Much like neural networks, they provide good-enough solution in short amount of time, but rarely find the best one. While they're not as popular as neural networks nor as widely used, they still have their place, as we can use them to solve complicated problems very fast, without expensive training rigs and with no knowledge of math.

Genetic algorithms can be used for variety of tasks, for example for determining the best radio antenna shape, aerodynamic shapes of cars and planes, wind mill shapes, or various queing problems. We'll use it to print "Hello, World!".

How does it work?

Genetic algorithm works in three steps.

1. Generate random solutions
2. Test how good they are
3. Pick the best ones, breed and mutate them, go to step 2

It works just like evolution in nature. First, we generate randomised solutions to our problem (in this case: random strings of letters).

Then, we test each solution and give it points, where better solutions gain more points. In our problem, we would give one point for each correct letter in the string.

Afterwards, we pick the best solutions and breed it together (just combine the strings). It's not bad idea to mutate (or randomize) the string a bit.

We collect the offsprings, and repeat the process until we find good enough solution.

Generate random solutions

First of all, we need to decide in which form we will encode our solutions. In this case, it will be simply string. If we wanted to build race cars, we would encode each solution (each car) as array of numbers, where first number would be size of the first wheel, the second number would be size of the second wheel, etc. If we wanted to build animals that try to find food, fight and survive, we would choose a decision tree (something like this).

So let's start and make few solutions, or entities. One hundred should be enough.

from random import randint

goal = "Hello, World!"
allowed_characters = list("qwertyuiopasdfghjklzxcvbnmQWERTYUIOPASDFGHJKLZXCVBNM ,!")

def get_random_entity(n, string_length):
    entities = []
    for _ in range(0, n):
        entity = ""
        for _ in range(0, string_length):
            entity += allowed_characters[randint(0, len(allowed_characters)-1)]
        entities.append(entity)
    return entities

print(get_random_entity(100, 13))

Test how good they are

This is called a "fitness function". Fitness function determines how good a solution is, be it a car (travel distance), animal (food gathered), or a string (number of correct letters).

The most simple function we can use right now will simply count correct letters. If we wanted, we could make something like Levenshtein distance instead.

def get_fitness(entity):
    points = 0
    for i in range(0, len(entity)):
        if goal[i] == entity[i]:
            points += 1
    return points

Crossover and mutation

Now it's time to select the best ones and throw away the less fortunate entities. Let's order entities by their fitness.

Crossover is a process, when we take two entities (strings) and breed them to create new one. For example, we could just give the offspring one part from one parent and another part from second parent.

There are many ways how to do this, and I encourage you to try multiple approaches when you will be doing something like this.

P:  AAAABBB|BCCCC
P:  DDDDEEE|FGGGG

F1: AAAABBB|FGGGG

Or we can just choose at random which letter will go from which parent, which works the best here. After we have the offsprint (F1), we should mutate it. What if we were unfortunate, and H (which we need for our Hello, World!) was not in any of the 100 entities? So we take the string and for each character of the string, there is a small chance to mutate it - change it at random.

F1:  ADDDEBEFGCGG
F1`: ADHDEBEFGCGG

And it's done. Now kill certain part of old population. I don't know which percentage is best, but I usually kill about 90% of old population. The 90% that we killed will be replaced by new offsprings.

There is just one more thing: which entities do we select for crossover? It isn't bad idea - and it generally works just fine - to just give better entities higher chance to breed.

def get_offspring(first_parent, second_parent, mutation_chance):
    new_entity = ""
    for i in range(0, len(first_parent)):
        if randint(0, 100) < mutation_chance:
            new_entity += allowed_characters[randint(0, len(allowed_characters)-1)]
        else:
            if randint(0, 1) == 0:
                new_entity += first_parent[i]
            else:
                new_entity += second_parent[i]
    return new_entity

When we add everything together, we get this output:

Generation 1, best score: 2 ::: QxZPjoptHfNgX
Generation 2, best score: 3 ::: XeNlTOQuAZjuZ
Generation 3, best score: 4 ::: weolTSQuoZjuK
Generation 4, best score: 5 ::: weTgnC uobNdJ
Generation 5, best score: 6 ::: weTvny uobldb
Generation 6, best score: 6 ::: HellSy mYbZdC
Generation 7, best score: 7 ::: selOoXBWoAKn!
Generation 8, best score: 8 ::: HeTloSoWYZlh!
Generation 9, best score: 8 ::: sellpX WobKd!
Generation 10, best score: 9 ::: welloq WobSdb
Generation 11, best score: 9 ::: selloc WoZjd!
Generation 12, best score: 10 ::: wellxX WoVld!
Generation 13, best score: 10 ::: welltX World!
Generation 14, best score: 10 ::: welltX World!
Generation 15, best score: 10 ::: welltX World!
Generation 16, best score: 11 ::: zellov Wobld!
Generation 17, best score: 11 ::: Hellty World!
Generation 18, best score: 11 ::: welloX World!
Generation 19, best score: 11 ::: welloX World!
Generation 20, best score: 11 ::: welloX World!
Generation 21, best score: 12 ::: welloX World!
Generation 22, best score: 12 ::: Helloy World!
Generation 23, best score: 12 ::: Helloy World!
Generation 24, best score: 12 ::: Helloy World!
Generation 25, best score: 12 ::: Helloy World!
Generation 26, best score: 12 ::: Helloy World!
Generation 27, best score: 12 ::: Helloy World!
Generation 28, best score: 12 ::: Helloy World!
Generation 29, best score: 12 ::: Helloy World!
Generation 30, best score: 12 ::: Helloy World!
Generation 31, best score: 12 ::: Helloy World!
Generation 32, best score: 12 ::: Helloy World!
Generation 33, best score: 12 ::: Helloy World!
Generation 34, best score: 13 ::: Helloy World!
Generation 35, best score: 13 ::: Hello, World!

As we can see, we find pretty good solution very fast, but it takes very long to find perfect solution. The complete code is here.

Maintaining diversity

When we solve difficult problems, it starts to be increasingly important to maintain diversity. When all your entities are basically the same (which happened in this example), it's difficult to find other solutions than those that are almost the same as the currently best one. There might be a much better solution, but we didn't find it, because all solutions that are different to currently best one are discarded. Solving this is the real challenge of genetic algorithms. One of the ideas is to boost diverse solutions in fitness function. So for every solution, we compute distance to the current best solutions and add bonus points for distance from it.