Geek's Code Corner

In Magic: the Gathering, a collectible card game, competitive players tend to gravitate towards a few dozen of the best decks made up out of a subset of all available cards. For instance in the Legacy format nearly all 18000 cards can be played, yet you’ll only see about 500 show up in tournaments with some cards (e.g. Brainstorm and Force of Will) showing up in > 50% of all high-ranking decks. When playing in such an event it is key to quickly identify your opponent’s deck and adapt your own game plan accordingly. Top players are able to very quickly do this, bad ones like myself need a few more turns. Here we’ll see if we can train a model that takes in a few known cards and outputs a prediction which deck is being played. In this blog-post I’ll show you how I made a classifier that can take a list of known cards in your opponent’s deck and return a list of possible decks they are playing. But let’s start with two examples of what it can do first. Imagine this scenario, on the first turn your opponent leads with Wasteland, on his second turn he plays Read more…

All our lives were impacted by COVID-19, but what was the impact on KeyForge sales? There was a very technical post last month, this one is intended for everyone that cares about KeyForge, but not the details of how to create models. 500 000 fewer decks registered during COVID This shouldn鈥檛 come as a surprise to anyone, during COVID, with the world going in lockdown there were fewer decks registered. With local game stores closed, in person games were hard if not impossible at times and the competitive scene died out entirely. So no decks are being purchased for sealed play and there was no need to buy a bunch of deck to find that one with the combo that could take you to the top of the leaderboards. With people unable to get together and play, they play less and the decks they do purchase will last longer as the novelty doesn鈥檛 wear off nearly as fast. The graph below shows, in blue, how many decks were registered in the master vault since the release up until this article was posted. The gray line indicates how many decks the model predicts would have been registered had COVID-19 never happened. Read more…

Effective insulation is key in modern homes, offering comfort and energy efficiency. However, it often comes with a need for better ventilation to maintain a healthy air quality. In this guide, we’re focusing on an innovative, cost-effective way to enhance your home’s air quality using some clever technology. We’ll show you how to combine the IKEA Vindriktning air quality monitor with an ESP8266 microcontroller and a BME680 Volatile Organic Compounds (VOC) sensor (which includes CO2). With this IKEA-hack, you’ll be able to make informed decisions about when to open windows, increase ventilation, or change filters, all based on accurate data. Plus, it integrates smoothly with smart home systems like Home Assistant using ESPHome. In this guide, we’re diving deep into every nook and cranny of this project. We’ll start from the ground up with the electronics, smoothly sail through the integration with Home Assistant (HA), and even sprinkle in some savvy tips on how to interpret the results. What you need? While this concept of enhancing the IKEA Vindriktning isn’t exactly breaking new ground – there’s already a plethora of information and various hacks out there – the advantage of being late to the party is that I can cherry-pick Read more…

In a previous post about clustering the palmer penguins using PyMC3, the mixture model was implemented in the model itself. Though this is great for understanding what is going on, it isn’t very efficient when sampling the model. Furthermore, observations (here penguins) were assigned to groups (here the species of penguin) using only one of the samples taken from the model, since we sample the model thousands of times it is a shame not to take all that data into account. When acquiring more observations, you typically don’t want to re-run the entire sampling step (here it takes minutes but this could be hours or days for more complex mixtures or larger datasets). Ideally you can fit the model on some initial data and then predict the group using new, previously unseen, data. With the code in the previous post this was not possible, in the code below we’ll explore how this can be done using PyMC3. So while we had a great model already, a few trick can still be used to further improve upon it. This GitHub repository contains all the code, which is based on the code discussed on the PyMC3 forums here and here. Both threads Read more…

How well do you know your algorithms? If you are a self-taught coder (like me), this might be a gap in your knowledge! This post is my attempt to get some extra practise under my belt writing algorithms to solve a few coding problems. On CodeForces, a site where there are programming competitions, the assignments from previous competitions are all available as well as the infrastructure to check your solution. So let’s use this resource to level up our programming a little. Getting Started On CodeForces all problems have a difficulty ranging from 800 to 3500. The lower the difficulty the easier to problem is to solve. To get started I recommend picking an easy problem to tackle to get familiar with the platform, how to get the input, how to submit a solution, … I picked problem 119A titled “Epic Game”, with the lowest difficulty of 800, to dip my toes in the water. The goal is to write a problem solve this issue: Simon and Antisimon play a game. Initially each player receives one fixed positive integer that doesn’t change throughout the game. Simon receives number a and Antisimon receives number b. They also have a heap of Read more…