Why are data scientists such bad coders? 🙈

Did you find some?

from * import *

@Demolishun ha. No.

I'm looking at one of our repos, is just a bit all over the place.

I've never seen code written by a data scientist and thought "yes, that's production ready".

Quality varies, I get it's not the primary skillset, but some of it is hard to read.

eg the python class with a list called "index" and another for the value. Who needs enumerate anyway?

And why bother with context managers?

Python 3? What's that?

They don't usually compare about long term maintainability as it's not their problem, so I don't think they usually put the thought in. It's also not really their area to be fair - writing clean, maintainable code takes experience & is kinda a skillset on its own.

Still, that's nothing compared to the code I've seen mathematicians hack together in Matlab. Blimey.

Haha, I work with data scientists and can confirm

True 100%, I started to despite Pandas because of how data scientists abuse the fuck out of it.

Better they be bad coders than bad data scientists I guess

You should see hardware engineers. They're as bad, if not worse.

@RememberMe hardware engineers as in VHDL? Coz most of my good coding practices come from VHDL hardware engineers.

@atheist I meant more like VLSI engineers - but I've met a fair number of HDL folks who were terrible at regular stuff. Also, fuck TCL and TCL scripts in general.

However I like how both VHDL and SystemVerilog are such primitive languages that you're pretty much forced to write good code or it doesn't scale at *all*.

Also there is nothing worse than a person who doesn't know HDL trying to design something in SystemVerilog...unlike VHDL that language is so easy to abuse (not that VHDL is all that much better tbh).

Why are qa engineers such bad coders ? Why are developers such bad sales people ? Why are sysadmins so bad at compassion ?

Why are Bananas such bad Apples ?
Catch my drift ?

@DarkMukke how many apples probably had bananaing lessons at uni?

AKA coding is usually part of the data science curriculum.

@atheist How many coders with degrees had coding subjects in their curriculum? Doesn't stop them from shitting all over the codebase

@atheist yeah I had English during my master's, still suck at it..

To be fair, I still prefer data scientist Python code over Python written by a Java dev. Speaking from experience.

@620hun worse than java written by a python dev?

@atheist I've seen very similar solutions in production at my previous job. The Java dev just made a class for literally everything you can imagine.

@DarkMukke

I think the issue is with the definition of "Data Scientist".

First of all, calling them scientists is ridiculous.

I think there are roughly three types:

1. Data statisticians — The person who validates existing assumptions using data, and performs rather complicated queries by request. Finds Python bothersome, prefers to use SQL directly, or R if they have to. Rigorously schooled statisticians who bitch about statistical significance and normalized data sets are most useful to a company.

2. Data-driven business analists — Someone who actively searches for patterns, even when there are none. Junior development skills, writes super messy PyTorch code. EagerSuperiors love this person: "wow, my machine learning model confirms your manager instincts exactly"

3. Data visualization designers — hates code, just wants to drag & drop stuff into colorful dashboards. Superiors love this person: Watching dashboards gives job security to managers.

@620hun lol what a nightmare. I love how the message is stored - because you never know when you might “need” it for later 🧐

Tell me more. If you see data scientist wrote CODE in MATLAB and call it a software and server .

@bittersweet does it apply to software 'engineer' too?

@curiousjoe

I think there are more parallels between software engineers and mechanical/civil engineers than there are parallels between data scientists and research chemists or experimental physicists.

A software engineer will largely go through the same steps as a "real" engineer. You get requirements and constraints. You make a technical design which is mindful of the cost, time and quality triangle. You build within certain tolerances, test your implementation rigorously.

A data scientist is usually not a scientist. Sure, they apply parts of the scientific method, but their job is usually more constrained to just analysis of existing data. The ones I've met would not get involved in the full cycle of research, hypothesis, (statistically significant) experimentation, conclusion, etc.

Maybe I'm biased, but I also feel like "science" implies something more fundamental, more generic, more academical than achievable within the context of most businesses.

@bittersweet really depends on who you ask I guess. I've encountered a whole bunch of scientists (data and otherwise) from companies who can science just as well as any "legit" scientist. The constraints of a corporate/company environment are just as legit constraints on research as any other.

@RememberMe Yeah but as a chemist I already feel like psychology isn't real science... Can't start awarding scientist badges to people who write SQL queries on sales orders.

@bittersweet that's a pretty low bar for "data scientist" I guess.

And as usual, knowing what to do with sales orders can be pretty complex too. That's one large component of recommendation engines for example.