|
raminasi posted:you sound like a pleasure to work with actually,
|
# ? Jun 4, 2016 01:37 |
|
|
# ? May 25, 2024 22:05 |
|
LOOK I AM A TURTLE posted:i guess youre referring to TransactionScope? its basically a way to use the same transaction in different methods, classes or even assemblies without having to explicitly pass around a transaction object or do manual commit/rollback. when you do using (var scope = new TransactionScope()) you either create a new transaction or enlist in the existing one if someone already created a scope above you in the stack. when you've finished doing your databasey thing you call scope.Complete() and dispose the scope object. If there are no remaining open scopes the transaction gets committed, if theres at least one scope that hasnt voted to commit yet, nothing happens for the time being. if a scope closes without having voted to commit (or if an exception is thrown inside the scope) the transaction is automatically rolled back. so it's basically a re-entrant lock?
|
# ? Jun 4, 2016 03:35 |
|
oh, my code only got a +1 in the review, what did i gently caress up? "your imports are in the wrong order, see this page in the wiki." he was right, there was a page in the wiki. that he had made. under "local environment setup" not "code style" and there was no announcement or discussion of the change in the developer channel. loving fine, i can change the settings in my ide, just not like this. also "it looks like you're not using this some.package.SomeThing import anywhere" "check lines 95...110, method setupSomeThing() " (i loving run "optimize imports" on commit, goddammit)
|
# ? Jun 4, 2016 06:24 |
|
goto-style error handling is an interesting idiom
|
# ? Jun 4, 2016 06:31 |
|
Wheany posted:oh, my code only got a +1 in the review, what did i gently caress up? condolences for working with a petty coding standard vigilante
|
# ? Jun 4, 2016 06:42 |
|
Strumpy posted:condolences for working with a petty coding standard vigilante it's fine, as long as i am made aware of the standards and i can discuss them earlier than in code review.
|
# ? Jun 4, 2016 07:03 |
|
Wheany posted:he was right, there was a page in the wiki. that he had made. i guess someone was bound to make the wiki eventually but it's still a shock when it happens
|
# ? Jun 4, 2016 09:55 |
|
Soricidus posted:
Our wiki has 3 different coding standards. Most people just roll a dice and go with whatever.
|
# ? Jun 4, 2016 12:46 |
|
Wheany posted:it's fine, as long as i am made aware of the standards and i can discuss them earlier than in code review. get thee a linter make a pre-commit hook wired up to that linter that rejects commits that don't pass and you'll get rid of vast majority of this don't waste your code review energy arguing about tabs
|
# ? Jun 4, 2016 14:13 |
|
someone got a lecture in our slack for using "fixing" in their commit summary and WASTING SPACE WITH USELESS TEXT goddammit
|
# ? Jun 4, 2016 14:16 |
|
every once in a while here someone presents about how to make nice linear git history "stories" and I am so glad i dont work with any of those people
|
# ? Jun 4, 2016 15:30 |
|
"lets browse git history that is older than HEAD~4" -no one, ever.
|
# ? Jun 4, 2016 15:43 |
|
Corla Plankun posted:"lets browse git history that is older than HEAD~4" randomly browse? no but run blame, oh yes
|
# ? Jun 4, 2016 15:51 |
|
Brain Candy posted:randomly browse? no also bisect
|
# ? Jun 4, 2016 16:04 |
|
Sagacity posted:At some point atlassian decided to remove the wiki markup editor from confluence and forced you to use the HTML editor because it was perfectly needs suiting, according to them. so they did not backpedal one bit they changed the internal storage format from markup-as-text-blob to xml tag-soup-as-text-blob. markup goes in, but xml tag soup comes out. you can still POST markup to confluence but you can'd edit any pages, so all markup producing/consuming tools are broken additionally, the xml soup is not meaningfully human-readable and doesn't preserve whitespace so basically there's no way to use confluence except through the web interface
|
# ? Jun 4, 2016 16:07 |
|
fritz posted:also bisect bisect doesn't require "linear stories" tho, just the occasional commit that actually compiles.
|
# ? Jun 4, 2016 16:08 |
|
Corla Plankun posted:every once in a while here someone presents about how to make nice linear git history "stories" and I am so glad i dont work with any of those people Corla Plankun posted:"lets browse git history that is older than HEAD~4" in theory, a sanitary git history is useful for chasing bugs with 'git bisect' in reality, nobody uses git bisect
|
# ? Jun 4, 2016 16:13 |
|
Notorious b.s.d. posted:in theory, a sanitary git history is useful for chasing bugs with 'git bisect' Them be fighting words. Although to be fair, I haven't used it either. My job used TFS over slow-rear end connection and emulating bisect when chasing old bug made me want to kill myself.
|
# ? Jun 4, 2016 16:22 |
|
one time i successfully used git bisect to locate the source of a regression bug ama
|
# ? Jun 4, 2016 16:27 |
|
I use git bisect maybe twice a year, and it's totally worth the "pain" of not being a gaping rear end in a top hat with your commits
|
# ? Jun 4, 2016 16:33 |
|
Notorious b.s.d. posted:in theory, a sanitary git history is useful for chasing bugs with 'git bisect' i've used bisection a whole bunch. it's really good and just one of many reasons why anyone who merges poo poo that doesn't compile into master should be shot.
|
# ? Jun 4, 2016 17:53 |
|
Notorious b.s.d. posted:in theory, a sanitary git history is useful for chasing bugs with 'git bisect' i used git bisect just this last week
|
# ? Jun 4, 2016 17:53 |
|
matrices are cool. i just refactored my garbage-nn project and because it didnt immediately crash due to a matrix dimension mismatch im pretty sure everything came through fine without having to wait a few minutes to see if the results still make sense its like type systems but for numbers although what was expected to be a few minutes turned out to be a few seconds, lol 25x speedup, thanks matrix math i should figure out how to get the math.net cuda bindings to work
|
# ? Jun 4, 2016 18:06 |
I've only used bisect once, but it lead me directly to a really really subtle bug in what I thought was a minor change that I had written a month ago, so it was totally worth it. I try to never commit things which don't compile so that I can bisect better now.
|
|
# ? Jun 4, 2016 18:17 |
|
interactive rebasing is super cool, what are you dorks complaining about
|
# ? Jun 4, 2016 19:17 |
|
Wheany posted:oh, my code only got a +1 in the review, what did i gently caress up? Good thing to know their priorities are in the right place
|
# ? Jun 4, 2016 19:40 |
|
i try to get my team to janitor their feature branch history at least a little because left to their own devices they merge 25 commits that just say "wip" and then one last one that says "works now" every time they do a ticket.
|
# ? Jun 4, 2016 19:40 |
|
i got the cuda bindings built and working but they're slower than whatever the default native provider was (openblas?)
|
# ? Jun 4, 2016 19:43 |
|
I treat commit messages as justification for the change, eg what does the change fix or improve Changes introduce risk, justify the risk
|
# ? Jun 4, 2016 19:49 |
|
jony neuemonic posted:i try to get my team to janitor their feature branch history at least a little because left to their own devices they merge 25 commits that just say "wip" and then one last one that says "works now" every time they do a ticket. i do this and clean it up to bullet points before i push the branch, but sometimes I have to merge in another branch mid-way and that breaks git reset so I have to push my poop to origin and feel bad
|
# ? Jun 4, 2016 19:49 |
|
jony neuemonic posted:i try to get my team to janitor their feature branch history at least a little because left to their own devices they merge 25 commits that just say "wip" and then one last one that says "works now" every time they do a ticket.
|
# ? Jun 4, 2016 22:30 |
|
but seriously good for you, like most things that poo poo programmers overlook it seems useless until it's suddenly super important why didn't anybody tell me this would happen guys
|
# ? Jun 4, 2016 23:07 |
|
Bloody posted:i got the cuda bindings built and working but they're slower than whatever the default native provider was (openblas?) how big's your data b/c you might be getting hit by data transfer overhead or something
|
# ? Jun 4, 2016 23:11 |
|
fritz posted:how big's your data b/c you might be getting hit by data transfer overhead or something its not super big and im relying on like a beta cuda provider to not pointlessly sling poo poo back and forth to the gpu for like no reason at all
|
# ? Jun 4, 2016 23:35 |
|
anyway my garbage neural network is down to around 2 seconds per training generation (from a minute recently and several minutes longer ago) and eventually gets up to around 94.9% classification rate on the mnist handwriting set which is p deece
|
# ? Jun 4, 2016 23:39 |
Bloody posted:anyway my garbage neural network is down to around 2 seconds per training generation (from a minute recently and several minutes longer ago) and eventually gets up to around 94.9% classification rate on the mnist handwriting set which is p deece This sounds really cool. Would you recommend http://neuralnetworksanddeeplearning.com? It seems like you are enjoying it, and I've been meaning to brush up on my ML skills.
|
|
# ? Jun 4, 2016 23:44 |
|
VikingofRock posted:This sounds really cool. Would you recommend http://neuralnetworksanddeeplearning.com? It seems like you are enjoying it, and I've been meaning to brush up on my ML skills. yep thats the exact book. i took the python code and rewrote it in c# and i've been mucking around with it loads to learn-by-messing around with
|
# ? Jun 4, 2016 23:57 |
Yeah you had mentioned that you were using that book upthread. I'll check it out!
|
|
# ? Jun 5, 2016 00:05 |
|
oh ya lol yeah id recommend it. it does things in a very mathsy sort of way which isnt my favorite but there's code provided and lots of words and im definitely learning
|
# ? Jun 5, 2016 00:14 |
|
|
# ? May 25, 2024 22:05 |
|
neuralnetworksanddeeplearning is a good book full of good stuff but i don't really like how it explains backprop. imo the right way to explain backprop is to draw lots of boxes with activations going forward and gradients going backward like in these slides http://www.cs.nyu.edu/~yann/2005s-V22-0480-006/diglib/lecture04a-backprop.pdf red is the computation graph of your neural network and green is gradients from backprop. drawing it like this isn't just a nice way to visualize it but it the backprop algorithm actually breaks apart into boxes just like in this picture so you can write a bit of code for each box and you can plug them together in whatever fancy graph structure you want and if you build the boxes right you get gradients of the full graph for free. this structure is pretty well known, but the only place i know of where someone has bothered to write it down this way is in really old yann lecun slides. if you poke around the slides for yanns old classes you can find lots of cool stuff like this http://www.cs.nyu.edu/~yann/2005s-V22-0480-006/schedule.html torch7 is built using this idea and its worth looking at how their modules are built if you want a concrete example of how to make boxes that fit together in a generic way.
|
# ? Jun 5, 2016 01:46 |