After listening to two of our senior devs play ping pong with a new member of our team for TWO DAYS!
DevA: "Try this.."
Junior: "Didn't work"
DevB: "Try that .."
Junior: "Still not working"

I ask..
Me:"What is the problem?"
Few ums...uhs..awkward seconds of silence

Junior: "App is really slow. Takes several seconds to launch and searching either crashes or takes a really long time."
DevA: "We've isolated the issue with Entity Framework. That application was written back when we used VS2010. Since that application isn't used very often, no one has had to update it since."
DevB: "Weird part is the app takes up over 3 gigs of ram. Its obviously a caching issue. We might have to open up a ticket with Microsoft."
Me: "Or remove EF and use ADO."
DevB: "That would be way too much work. The app is supposed to be fully deprecated and replaced this year."
Me: "Three of you for the past two days seems like a lot of work. If EF is the problem, you remove EF."
DevA: "The solution is way too complicated for that. There are 5 projects and 3 of those have circular dependencies. Its a mess."
DevB: "No fracking kidding...if it were written correctly the first time. There aren't even any fracking tests."
Me:"Pretty sure there are only two tables involved, maybe 3 stored procedures. A simple CRUD app like this should be fairly straight forward."
DevB: "Can't re-write the application, company won't allow it. A redesign of this magnitute could take months. If we can't fix the LINQ query, we'll going to have the DBAs change the structures to make the application faster. I don't see any other way."

Holy frack...he didn't just say that.

Over my lunch hour, I strip down the WPF application to the basics (too much to write about, but the included projects only had one or two files), and created an integration test for refactoring the data access to use ADO. After all the tests and EF removed, the app starts up instantly and searches are also instant. Didn't click through all the UI, but the basics worked.

Sat with Junior, pointed out my changes (the 'why' behind the 'what') ...and he how he could write unit tests around the ViewModel behavior in the UI (and making any changes to the data access as needed).

Today's standup:
Junior: "Employee app is fixed. Had some help removing Entity Framework and how it starts up fast and and searches are instant. Going to write unit tests today to verify the UI behaivor. I'll be able to deploy the application tomorrow."
DevA: "What?! No way! You did all that yesterday?"
Me: "I removed the Entity Framework over my lunch hour. Like I said, its basic CRUD and mostly in stored procedures. All the data points are covered by integration tests, but didn't have time for the unit tests. It's likely I broke some UI behavior, but the unit tests should catch those."
DevB: "I was going to do that today. I knew taking out Entity Framework wouldn't be a big deal."

Holy fracking frack. You fracking lying SOB. Deeeep breath...ahhh...thanks devRant. Flame thrower event diverted.

Let the student use their own laptops. Even buy them one instead of having computers on site that no one uses for coding but only for some multiple choice tests and to browse Facebook.

Teach them 10 finger typing. (Don't be too strict and allow for personal preferences.)

Teach them text navigation and editing shortcuts. They should be able to scroll per page, jump to the beginning or end of the line or jump word by word. (I am not talking vi bindings or emacs magic.) And no, key repeat is an antifeature.

Teach them VCS before their first group assignment. Let's be honest, VCS means git nowadays. Yet teach them git != GitHub.

Teach git through the command line. They are allowed to use a gui once they aren't afraid to resolve a merge conflict or to rebase their feature branch against master. Just committing and pushing is not enough.

Teach them test-driven development ASAP. You can even give them assignments with a codebase of failing tests and their job is to make them pass in the beginning. Later require them to write tests themselves.

Don't teach the language, teach concepts. (No, if else and for loops aren't concepts you god-damn amateur! That's just syntax!)

When teaching object oriented programming, I'd smack you if do inane examples with vehicles, cars, bikes and a Mercedes Benz. Or animal, cat and dog for that matter. (I came from a self-taught imperative background. Those examples obfuscate more than they help.) Also, inheritance is overrated in oop teachings.

Functional programming concepts should be taught earlier as its concepts of avoiding side effects and pure functions can benefit even oop code bases. (Also great way to introduce testing, as pure functions take certain inputs and produce one output.)

Focus on one language in the beginning, it need not be Java, but don't confuse students with Java, Python and Ruby in their first year. (Bonus point if the language supports both oop and functional programming.)

And for the love of gawd: let them have a strictly typed language. Why would you teach with JavaScript!?

Use industry standards. Notepad, atom and eclipse might be open source and free; yet JetBrains community editions still best them.

For grades, don't your dare demand for them to write code on paper. (Pseudocode is fine.)

Don't let your students play compiler in their heads. It's not their job to know exactly what exception will be thrown by your contrived example. That's the compilers job to complain about. Rather teach them how to find solutions to these errors.

Teach them advanced google searches.

Teach them how to write a issue for a library on GitHub and similar sites.

Teach them how to ask a good stackoverflow question :>

Ok so a very quick background: I didn't get a job until I needed one after my phone broke and I didn't have the money to buy another one. (I'm a student still for those who don't know lol.)

>> Phone randomly breaks.
>> Don't have the money for a new one
>> Searches for low skill jobs (ie cashier) that would work with me in terms of how many hours I work and whenever those hours are.
>> Apply to like 15 (not even exaggerating either lol) jobs
>> Wait for responses
>> One day goes by. Nothing
>> Two days go by. Nothing
>> Three days go by. Nothing
>> Fourth day rolls around and I get a call about one. I answer, tell him I'm available starting that Monday.
>> (Keep in mind I'm on an old temporary phone)
>> Next day I buy a new phone (didn't have to pay anything up front aside from the taxes on it, as it's on a payment plan)
>> Reset old phone after usage
>> Monday rolls around and I drive to the location of the job, and walk in the door asking for x.
>> "I'm sorry sir, who? We don't have anyone here by that name."
>> I panic and hop in my car, and try to find the address of the store I applied to. I find out it's different than where I went.
>> Start driving there and call that phone number. I ask to speak to x.
>> "I'm sorry sir, there's nobody here by that name."
>> Call literally every other location in the city and ask for x, but nothing.
>> Since I'm already on the way, I drive to the location of the store I had applied to. Whenever I get there, the manager spends half an hour on the phone trying to figure out where I belong. Nothing.
>> By this point, it's well over an hour past whenever I was supposed to show up, so I gave up because I figured they probably wouldn't have hired me anyways.
>> I get home determined af to figure out who the hell called me.
>> I remembered that Verizon has call logs you can look at online.
>> I go back through it and find the number. Google it.

Here's where the story gets a lil funny now.

>> Number shows up for a store that I applied to who's name sounds a LOT like the first store.
>> Called them and explained what happened, and told them I'd be there asap if they still wanted me.
>> That was like 6 months ago and I'm still here lol

When you search for 'Microsoft startup program' and it only returns articles that show you how to add or remove startup programs from windows.

*sigh*

The next step for improving large language models (if not diffusion) is hot-encoding.

The idea is pretty straightforward:
Generate many prompts, or take many prompts as a training and validation set. Do partial inference, and find the intersection of best overall performance with least computation.

Then save the state of the network during partial inference, and use that for all subsequent inferences. Sort of like LoRa, but for inference, instead of fine-tuning.

Inference, after-all, is what matters. And there has to be some subset of prompt-based initializations of a network, that perform, regardless of the prompt, (generally) as well as a full inference step.

Likewise with diffusion, there likely exists some priors (based on the training data) that speed up reconstruction or lower the network loss, allowing us to substitute a 'snapshot' that has the correct distribution, without necessarily performing a full generation.

Another idea I had was 'semantic centering' instead of regional image labelling. The idea is to find some patch of an object within an image, and ask, for all such patches that belong to an object, what best describes the object? if it were a dog, what patch of the image is "most dog-like" etc. I could see it as being much closer to how the human brain quickly identifies objects by short-cuts. The size of such patches could be adjusted to minimize the cross-entropy of classification relative to the tested size of each patch (pixel-sized patches for example might lead to too high a training loss). Of course it might allow us to do a scattershot 'at a glance' type lookup of potential image contents, even if you get multiple categories for a single pixel, it greatly narrows the total span of categories you need to do subsequent searches for.

In other news I'm starting a new ML blackbook for various ideas. Old one is mostly outdated now, and I think I scanned it (and since buried it somewhere amongst my ten thousand other files like a digital hoarder) and lost it.

I have some other 'low-hanging fruit' type ideas for improving existing and emerging models but I'll save those for another time.

Data wrangling is messy

I'm doing the vegetation maps for the game today, maybe rivers if it all goes smoothly.
I could probably do it by hand, but theres something like 60-70 ecoregions to chart,
each with their own species, both fauna and flora. And each has an elevation range its
found at in real life, so I want to use the heightmap to dictate that. Who has time for that? It's a lot of manual work.

And the night prior I'm thinking "oh this will be easy."

yeah, no.

(Also why does Devrant have to mangle my line breaks? -_-)

Laid out the requirements, how I could go about it, and the more I look the more involved
it gets.

So what I think I'll do is automate it. I already automated some of the map extraction, so
I don't see why I shouldn't just go the distance.
Also it means, later on, when I have access to better, higher resolution geographic data, updating it will be a smoother process. And even though I'm only interested in flora at the moment, theres no reason I can't reuse the same system to extract fauna information.
Of course in-game design there are some things you'll want to fudge. When the players are exploring outside the rockies in a mountainous area, maybe I still want to spawn the occasional mountain lion as a mid-tier enemy, even though our survivor might be outside the cats natural habitat. This could even be the prelude to a task you have to do, go take care of a dangerous
creature outside its normal hunting range. And who knows why it is there? Wild fire? Hunted by something *more* dangerous? Poaching? Maybe a nuke plant exploded and drove all the wildlife from an adjoining region?
who knows.

Having the extraction mostly automated goes a long way to updating those lists down the road.

But for now, flora.

For deciding plants and other features of the terrain what I can do is:

* rewrite pixeltile to take file names as input,
* along with a series of colors as a key (which are put into a SET to check each pixel against)
* input each region, one at a time, as the key, and the heightmap as the source image
* output only the region in the heightmap that corresponds to the ecoregion in the key.
* write a function to extract the palette from the outputted heightmap. (is this really needed?)
* arrange colors on the bottom or side of the image by hand, along with (in text) the elevation in feet for reference.

For automating this entire process I can go one step further:
* Do this entire process with the key colors I already snagged by hand, outputting region IDs as the file names.
* setup selenium
* selenium opens a link related to each elevation-map of a specific biome, and saves the text links
(so I dont have to hand-open them)
* I'll save the species and text by hand (assuming elevation data isn't listed)
* once I have a list of species and other details, to save them to csv, or json, or another format
* I save the list of species as csv or json or another format.
* then selenium opens this list, opens wikipedia for each, one at a time, and searches the text for elevation
* selenium saves out the species name (or an "unknown") for the species, and elevation, to a text file, along with the biome ID, and maybe the elevation code (from the heightmap) as a number or a color (probably a number, simplifies changing the heightmap later on)

Having done all this, I can start to assign species types, specific world tiles. The outputs for each region act as reference.

The only problem with the existing biome map (you can see it below, its ugly) is that it has a lot of "inbetween" colors. Theres a few things I can do here. I can treat those as a "mixing" between regions, dictating the chance of one biome's plants or the other's spawning. This seems a little complicated and dependent on a scraped together standard rather than actual data. So I'm thinking instead what I'll do is I'll implement biome transitions in code, which makes more sense, and decouples it from relying on the underlaying data. also prevents species and terrain from generating in say, towns on the borders of region, where certain plants or terrain features would be unnatural. Part of what makes an ecoregion unique is that geography has lead to relative isolation and evolutionary development of each region (usually thanks to mountains, rivers, and large impassible expanses like deserts).

Maybe I'll stuff it all into a giant bson file or maybe sqlite. Don't know yet.

As an entry level programmer I may not know what I'm doing, and I may be supposed to be looking for a job, but that won't stop me from procrastinating.

Data wrangling is fun.

I just started a new job last week. Old-school sysadmin role for a pretty old-school company, but the pay is nice and the kids've gotta eat.

They gave me a windows laptop. I haven't used windows for work or as a daily driver since 2016, and now, a week into trying to make this machine work for me, I have the following observations to report.

WSL is nice. It's nice to have it installed(though actually installing it was an adventure unto itself), and to set alacritty to open my default user prompt straight into that is very nice. As terminal emulators are by far my most used piece of software, that's nice to have.

Command-line software management through powershell, winget, and chocolatey are also very nice.

I like the accessibility offered by autohotkey, though there is something of a learning curve on it. Once I get better with it, I suspect that what follows will be largely mitigated.

The Bad:
In general, Windows is janky. It feels like it's all kinda taped together without any particular cohesion in mind. As a desktop, it feels decidedly amateur, compared to the feature-mountain polish of MacOS, and especially compared to the flexibility and infinite possibilities of Linux.

Lots of screen real estate is wasted, with window decorations, and fonts that look terrible at smaller sizes, because the antialiasing of fonts is just terrible. Almost all the features I depend on in other desktops: ad-hoc searches and launches(alfred, rofi) are-- again --janky. They work, but they typically require more typing than alfred or rofi. I admit I haven't spent weeks on this problem yet, but I haven't found a workable solution yet with wox, hain, and keypirinha. Quick searches like what you get with alfred, alfred workflows, and the swiss army knife that is rofi, just aren't possible or reliable with the tools I've used so far, and most require some kind of indexing agent to fully function.

It beggars imagination that a desktop in which users are subjected to "default apps" that is purported to be acceptable for enterprise, professional use, does not have a default entry for text editor. I installed nvim-qt, and I want to use it to edit anything and everything I ever edit with text, but all too often, apps have hard-coded instructions to open text files with notepad.

I want to open certain URLs with firefox, certain ones with firefox developer edition, and others with vivaldi, and yet there is not an app available that I have seen yet in my searches that allows me to set this kind of configuration. I found one that's supposed to, but it just ignores everything I put into its config, and just opens MS Edge for everything. Jank.

Simple things take too long. Like the delay between when I laboriously hit ctrl-alt-del to bring up the login and when the actual text field appears, and the delay between that and when I want to start using the computer.

Changing some settings requires a reboot. Updating some software requires a reboot. Updating permissions on something sometimes requires a reboot. And those are all on top of the frequent requests to reboot for updates.

I would have thought Windows would have overcome most of the issues that create these problems, but it's just, as I said, amateur.

Control your searches like an ADULT damn it!!!

So we have records that can have any of a bazillion different reference numbers associated with them. No big deal. Everyone does right?

Our customer's love to run reports and so we have this one option for "just look at a hell of a lot of reference numbers". I call it the 'fuck all' search.

Really it is just there to find something that you don't know where a rando string or number might be in the record and just want to do a "fuck all" search across a number of likely fields to find it... and then presumably you'd be an adult and refine your search from there. LOL yeah right...

Customers get lazy and include that stupid option in their reports and we get a lot of.

Customer: "I always run this report (that includes the fuck all search) and now it isn't working. I want records that have ID 2222."

Me: "Yeah well that was only working because you were rando typing '2222' in like several fields and it would find those .... but now you quit doing that so it won't find them. If you want ID 2222, click the drop down and search by 'ID'. That will find it right away."

Customer: "But I want to just search by 'fuck all search' to find it..."

Me: "But then you get all these other records too right?"

Customer: "Yeah but I just delete them out of the spreadsheet ... "

Me: "Look watch this <screen share> there, look all records with an ID of 2222 and no more extra records you need to delete!!! How great is that?"

Customer: "But why do I have to do it this way now, I want to do it the old way..."

ಠ_ಠ

(granted I could add their ID to the fuck all search but we try to avoid adding too much because it gets out of hand / stops being useful the more fuck all it gets)