In Soviet Russia, QA systems ask YOU

Autor: -
Poslední dobou dělám samé meniální práce. V jistém smyslu mi to i vyhovuje, ale blbě se tím vytahuje. Teď jsem musel ručně zpracovat 542 otázek. Přišlo mi vtipné, že se tentokrát ptal systém mě.

YodaQA ke každé otázce vygeneruje několik clues (předmět otázky, sousloví) a pokusí se v databázi najít k nim náležící koncepty. Například když se ptáte na "Who directed X-Men: Days of Future Past", tak mimo jiné najde koncept "X-Men: Days of Future Past". Nebo nenajde. A pak to je průser. Mým nejnovějším úkolem (zahrnutým v pojmu entity linking) bylo projít nějaký dataset (který jsem předtím vytvořil..) a tam ručně přiřadit koncepty. Občas to bylo lehký, protože ty koncepty vybral správně. Jindy jich namatchoval tucet, protože na jména bez příjmení to najde spoustu mrdek. Nejhorší bylo, když to nenašlo nic, protože ručně hledat koncepty je docela peklo. Musel jsem kvůli tomu pracovat s jazykem SPARQL, který slušně řečeno nedává smysl.

Zpočátku jsem to dělal v textovém editoru rychlostí cca 1 otázka za minutu. Mentor mi navrhl udělat webové rozhraní, kde by se klikalo. Docela jsem se tomu bránil, protože nemusím html, ale po 3 hodinách jsem v pythonu slepil něco základního. Samozřejmě jsem kvůli tomu musel udělat několik dalších skriptů a díky tomu se to protáhlo na dva dny místo jednoho odpoledne, ale zvládl jsem nakonec udělat otázku za 30s.

To je fajn, když musíte naprogramovat nástroje pro levnou pracovní sílu jenom, abyste ty samé nástroje museli používat. Dost mě to motivovalo si s tím hrát, aby to bylo pohodlnější.

Komentáře

Okomentovat

Populární příspěvky z tohoto blogu

Creating my own Point of Sale system, Part 0: Start

Autor: -
Obrázek
I had several topics I wanted to write about (both from my dabblings in machine learning and travel stories), but a more urgent thing came up. With a change in Czech legislature, all shopkeepers are now required to authorize all transactions at the Ministry of Finances' servers. They call it EET - Elektronicka Evidence Trzeb. (electronic records of sales) Sadly, POS  just reminds me of part-of-speech tags. To deal with the ambiguity, I'll just refer to it as EET system, even though it contains more. So what exactly is EET? From march on, every time we make a sale, we have to contact their servers and send them the amount with our hashed ID we got from them by registering. Afterwards, they send an unique receipt ID back, which we have to print and give to the customer. The customer can then at any point check the receipt on their website. The specifications are all published and the state doesn't require any license to use it. It's rather annoying and requires c
Další informace

Creating my own Point of Sale system, Part 1: Design

Autor: -
This took some time, I'm (mostly) back on track. When I chatted with my SI friends, the first thing I noticed was they asked a fuckton of questions. They grilled me about specifications, use cases, reliability or scalability. They also gave insane price quotes but I'll skip over that. The reason is simple: Specifications save time. I am definitely not used to that. I'm used to quickly hacking together a sqlite database for a chatbot in python and iterating bugfixes/feature additions as I go. So I thought, "Why not do it properly for once?" Well, that was also a mistake. The first thing one notices when starting a programming project is the sheer number of decisions. From the programming language to data types, databases and used design patterns, each choice has several impacts and you don't even know the right use case if you don't work with them for some time. So let's start with the language. I am quite fluent with Python (as in, I can solve mo
Další informace

Asian Squad

Autor: -
Obrázek
One of the things I always had problems with was getting along with fellow Vietnamese. At some point I actually hated them. It's hard to name the main reason, so I'll try to talk around it instead. We start with the context. Many Vietnamese (at least the ones I know) came to the Czech Republic from Russia after the split of Czechoslovakia, because it was relatively easy to slip though the borders. They mostly had a similar background - not particularly educated, but with a ruthless (they are fucking brutal  and will stop at nothing to succeed) business sense and a desire for financial security. The main income was selling fake  brand clothes at the market, smuggling other Vietnamese and tax evasion. My parents sold vegetables and fruit, which was at the time kinda good (after 10 years, we had our own flat, even though we came with 1000$), but could not compare to clothes. As a comparison, there was a family who lives next to us that arrived at the same time with the same amount
Další informace