|
What industry actually uses COCOMO in modern practice? I learned about it in my software engineering course in college in 2001, taught by a guy who hadn't been in industry for years, then never heard about it again.
|
#
?
Mar 10, 2017 16:26
|
|
|
|
| # ? May 12, 2024 02:11 |
|
|
csammis posted:What industry actually uses COCOMO in modern practice? I learned about it in my software engineering course in college in 2001, taught by a guy who hadn't been in industry for years, then never heard about it again. Nobody, because it's a piece of poo poo (is lies: schedule critical big enterprisey bullshit where they have dozens of architecture weirdos who don't actually code jesus christ what a piece of poo poo) The goal is to make it not a piece of poo poo
|
|
#
?
Mar 10, 2017 16:27
|
|
|
Well why polish the turd and compare it to COCOMO at all unless it's a form of COCOMO? Like Vulture Culture said that's just going to attract terrible managers and make people who understand software estimation even a little not take you seriously. If you've created a And yeah of course machine learning is going to have a hard time determining when a Github issue is closed, the halting problem hasn't been solved 
|
|
#
?
Mar 10, 2017 16:31
|
|
|
Yeah, I was thinking about COCOMO and how it's such a goddamn piece of poo poo while making it I'll call it Comestimate because I am very hungry right now and it'll also support JIRA when I actually release it Most actual practical machine learning problems attack a lovely version of the halting problem, because correlational structure expands in time indefinitely. Long-temporal-distance correlation is a fundamental feature of human language for example and long spatial distance correlation is a fundamental feature of all interesting image data. Think of 2nd-order phase transition where correlational distance expands to the whole object with infinite distance... but modern machine learning poo poo can attack it not too badly. Just like how modern actual halting can be detected a little bit less crappily by a few heuristics
|
|
#
?
Mar 10, 2017 16:38
|
|
|
Vulture Culture posted:My instinct is that you've created a tool that will allow terrible managers to make even worse decisions Is there anything that this thing can do so I can convince folks that it won't do that? (or convince managers to not do terrible things, but that's a lost cause)
|
|
#
?
Mar 10, 2017 16:48
|
|
|
curufinor posted:I'll call it Comestimate because I am very hungry right now Uh okay quote:Most actual practical machine learning problems attack a lovely version of the halting problem, because correlational structure expands in time indefinitely. Long-temporal-distance correlation is a fundamental feature of human language for example and long spatial distance correlation is a fundamental feature of all interesting image data. Think of 2nd-order phase transition where correlational distance expands to the whole object with infinite distance... but modern machine learning poo poo can attack it not too badly. Okay. Just to make sure I'm understanding what you want here, you want to validate the accuracy of your algorithms by recruiting a human person to look at the same Github project as your tool to estimate TTC on the issues then wait to see who is more right? That doesn't sound very practical unless it's a project with a high rate of issue creation and resolution which probably means it's intractable (or just boring) for a human to crawl all that data and come up with an estimate. How much are you paying for such a recruit?
|
|
#
?
Mar 10, 2017 16:48
|
|
|
I was planning to have a sample of like 25 issues that a bunch of folks would look at, to get less uneven data. There's no central limit theorem in this sort of thing (number of issues and predictability are deffo more fat-tailed than would be compatible with central limit theorem), but it should still get a little more accurate with more people, and it's not like you can get a bunch of Amazon Turkers to do this sort of thing. I guess I can try to find a venue where I can get a bunch of programmers to do 15 minute tasks but that's a whole can of worms.
|
|
#
?
Mar 10, 2017 16:57
|
|
|
curufinor posted:Comestimate ew
|
|
#
?
Mar 10, 2017 16:58
|
|
|
curufinor posted:Is there anything that this thing can do so I can convince folks that it won't do that?
|
|
#
?
Mar 10, 2017 16:59
|
|
|
It's basically a LSTM with like 20 weird tricks cribbed from the recesses of random papers I read over the last decade and a lot of weird issue tracker-specific features
|
|
#
?
Mar 10, 2017 17:01
|
|
|
curufinor posted:I'll call it Cum estimate because I am very hungry right now 
|
|
#
?
Mar 10, 2017 17:05
|
|
|
Ok, different name time, thank you
|
|
#
?
Mar 10, 2017 17:06
|
|
|
Come Estimate My Lord. Estimate Discreetly.
|
|
#
?
Mar 10, 2017 17:07
|
|
|
HardDiskD posted:Issues? On GitHub? Infinite time. Where do I collect my paycheck? lol now that I read this, this is basically literally true issue times on github are fat-tailed and close to pareto distributed, I think difficult to actually tease out differences from lognormal and pareto distributed value but it could go either way if it's pareto-distributed average times will literally tend towards infinite in theoretical expected value of course it still remains empirically possible to estimate, you just have to shove in a lot more information
|
|
#
?
Mar 10, 2017 17:10
|
|
|
curufinor posted:It's basically a LSTM with like 20 weird tricks cribbed from the recesses of random papers I read over the last decade and a lot of weird issue tracker-specific features Unfortunately spouting off machine learning heuristics and statistical distribution models is not a very good sales pitch for why anyone, particularly project management professionals, should care about this tool. Will this tool generate a prediction for time to completion for a completely novel task on a given project that has been analyzed? What about a completely novel project? How much data does it need to bootstrap? How does it estimate resources allocated to a task or is it agnostic of this? If the estimates are agnostic of assigned resources, how can it generate predictive data that's worth anything? How have you proven its correctness to date? edit: looking back I realize this sounds adversarial and if it comes off as being too much of an rear end in a top hat I apologize. The fact is if you want to introduce a new estimation tool into the marketplace there are going to be some tough questions to answer, like "can the market support yet another estimation tool?" and "is there even a market for this kind of estimation?" If this is a personal project then the stuff about the marketplace doesn't apply. Personal projects are great! If you use it as a portfolio project though, you may want to be prepared to answer the above list of questions anyway. I know I'd ask them if this was presented to me by a candidate. csammis fucked around with this message at 21:27 on Mar 10, 2017 |
|
#
?
Mar 10, 2017 20:46
|
|
|
csammis posted:Unfortunately spouting off machine learning heuristics and statistical distribution models is not a very good sales pitch for why anyone, particularly project management professionals, should care about this tool. Yes. If and only if the task comes in an issue. quote:What about a completely novel project? Nope. quote:How much data does it need to bootstrap? About 100 issues, empirically: I've been trying to get that to a lower number but it satisfices for right now quote:How does it estimate resources allocated to a task or is it agnostic of this? Completely agnostic, actually quote:If the estimates are agnostic of assigned resources, how can it generate predictive data that's worth anything? How you talk about the issue (a strangely lame predictor), the time series of how many issues people have been bumping up lately, the identities of creators and assignees, other weird facts about the issue. You can actually get a lot about the attitude of the issue from NLP stuff on followups, too. Weird-rear end features, basically. quote:How have you proven its correctness to date? Backtesting, just like stock market predictors, just on open source data. You may claim open source data is completely different from enterprisey sorts of thing, which may or may not be true. I've also been questing for enterprise data, but that's obviously less easy to get ahold of. And it would still be useful if it works on open source. I actually have no idea if the current model is incredible or not, just from having no human benchmarks. Software estimation, of course, is hard as hell for humans. But many of the things that make it hard (schedule pressure, panicking, assholeness of boss, hierarchical decision making) are uniquely human weaknesses. The ML model certainly does a fair bit better than random. quote:edit: looking back I realize this sounds adversarial and if it comes off as being too much of an rear end in a top hat I apologize. The fact is if you want to introduce a new estimation tool into the marketplace there are going to be some tough questions to answer, like "can the market support yet another estimation tool?" and "is there even a market for this kind of estimation?" If this is a personal project then the stuff about the marketplace doesn't apply. Personal projects are great! If you use it as a portfolio project though, you may want to be prepared to answer the above list of questions anyway. I know I'd ask them if this was presented to me by a candidate. I don't actually know if it'll be a personal project or an actual thing, yet. I have no idea for either of the two questions. But I've used many of the paid estimation tools and they suck so much, and cost so much and are so awful.... curufinor fucked around with this message at 21:36 on Mar 10, 2017 |
|
#
?
Mar 10, 2017 21:29
|
|
|
Just saw the plan for our post-remodel office layout. Current situation: rows of cubicles with high cube walls at our backs. Noise is a problem sometimes but we don't hear every employee in the office all of the time. I'm sure everyone can make a good guess as to what the new plan involves (hint: the word "collaborative" features prominently). 
|
|
#
?
Mar 11, 2017 01:48
|
|
|
Che Delilas posted:Just saw the plan for our post-remodel office layout.
|
|
#
?
Mar 11, 2017 01:54
|
|
|
Vulture Culture posted:At one of my previous jobs I was bothered enough by the proximity to my coworker's conversations that I built walls for my desk out of multi-ply corrugated cardboard and management never had the balls to ask me to take it down Last job, at a tiny little startup, I expensed a box to sit in with laptop. Model name? Gaylord Definitely sufficiently cozy esp if you line it with blankets Do not try if your company is more than 5 people
|
|
#
?
Mar 11, 2017 02:19
|
|
|
curufinor posted:Last job, at a tiny little startup, I expensed a box to sit in with laptop. The whole point is that the cat sits in the box, not you
|
|
#
?
Mar 11, 2017 02:24
|
|
|
Vulture Culture posted:At one of my previous jobs I was bothered enough by the proximity to my coworker's conversations that I built walls for my desk out of multi-ply corrugated cardboard and management never had the balls to ask me to take it down They're also putting us right next to customer support, because when they can't figure something out it's too much of a burden to walk 50 feet to the other side of the building to talk to one of the devs (or, you know, send an email or talk to us via Slack which we all have). What's that? Constant ringing phones with no walls or distance to mitigate the sound at all are a nightmare for the devs? gently caress YOU. They actually pushed for everyone using a single monitor because "more monitors isolate you." Dev department head had to kick and scream to get them to back off on that so we probably aren't even going to be able to arrange ourselves in own section the way we want. Shoulder-to-shoulder in the center of the room, basically, and if they don't let us rearrange it ourselves then I'm probably just gone at that point. There's no reason to ignore the devs' input on the layout (especially in our own section, jesus gently caress) this completely if they want us to do actual work. I'm more than half convinced it's pure theater, they're going to bring in some money that thinks they know what a small development company should look like and try to get us bought. God I'm so pissed. Che Delilas fucked around with this message at 05:58 on Mar 11, 2017 |
|
#
?
Mar 11, 2017 05:49
|
|
|
Che Delilas posted:They're also putting us right next to customer support, because when they can't figure something out it's too much of a burden to walk 50 feet to the other side of the building to talk to one of the devs (or, you know, send an email or talk to us via Slack which we all have). What's that? Constant ringing phones with no walls or distance to mitigate the sound at all are a nightmare for the devs? gently caress YOU. Eject! Eject!
|
|
#
?
Mar 11, 2017 16:13
|
|
|
EDIT: Nevermind, just found the newbie thread. Taking this there.
LanceHunter fucked around with this message at 19:28 on Mar 11, 2017 |
|
#
?
Mar 11, 2017 19:16
|
|
|
Volmarias posted:Eject! Eject! They've said it's "not set in stone" yet, and it's at least half a year away, so there may be time for me to conduct a campaign to get them to pull their heads out. I may have to resort to linking to Forbes articles since they don't seem to want to listen to the people that are actually creating the products they sell, but I'm going to give it a try. But if they keep being stupid and insist on taking away a reason I have to put up with substandard compensation? Yeah, I'm out.
|
|
#
?
Mar 11, 2017 22:13
|
|
|
Che Delilas posted:They've said it's "not set in stone" yet, and it's at least half a year away, so there may be time for me to conduct a campaign to get them to pull their heads out. I may have to resort to linking to Forbes articles since they don't seem to want to listen to the people that are actually creating the products they sell, but I'm going to give it a try. It's nice they're giving you 6 months to find a job like that...
|
|
#
?
Mar 12, 2017 00:01
|
|
|
Che Delilas posted:They've said it's "not set in stone" yet, and it's at least half a year away, so there may be time for me to conduct a campaign to get them to pull their heads out. I may have to resort to linking to Forbes articles since they don't seem to want to listen to the people that are actually creating the products they sell, but I'm going to give it a try. What's so great about this company versus others that you want to stay after they pull dumb moves like this?
|
|
#
?
Mar 12, 2017 00:03
|
|
|
piratepilates posted:What's so great about this company versus others that you want to stay after they pull dumb moves like this? https://en.wikipedia.org/wiki/Stockholm_syndrome The ifs will keep evolving. I've been there.
|
|
#
?
Mar 12, 2017 01:17
|
|
|
piratepilates posted:What's so great about this company versus others that you want to stay after they pull dumb moves like this? Right now? It's interesting work, they're flexible about hours (meaning both that we can start & end the day when we want AND they don't bitch about a less-than-40-hour-workweek), we have choices about what we work on (meaning who takes which feature and we can decide to inject our own work into the schedule if it's something we need to do or will help us), all of my co-workers including my direct boss are pretty great people, we can work from home when we feel like it, and the office is an environment that's generally conducive to getting poo poo done. The downsides are substandard compensation and the fact that the product owner is on the other side of the country and very disconnected from the team, which causes minor annoying problems but nothing we can't ignore or work around. Right now the balance of good vs. bad is such that rolling the dice on a new place feels like a pretty low-percentage move on my part, unless another company offered me a stellar compensation package (which hasn't been on the table for the opportunities that have come my way). Now, if my current place does this dumb poo poo with the office layout (which not only makes our lives miserable but sends a strong message about either how little they value the developers, or as I said before that they want to put on a play to potential acquirers), that would tip the scales quite a bit. Hughlander posted:It's nice they're giving you 6 months to find a job like that... I saw the plan at like 5:00 yesterday, so next week I'll be getting some more detail about how much their hearts are set on sabotaging dev productivity. It shouldn't take much for me to figure it out, and if it's clear there's no improving the coming situation then yeah, my plan is to start sending out applications next weekend and have a nice low-pressure job hunt.
|
|
#
?
Mar 12, 2017 07:29
|
|
|
Che Delilas posted:Right now? It's interesting work, they're flexible about hours (meaning both that we can start & end the day when we want AND they don't bitch about a less-than-40-hour-workweek), we have choices about what we work on (meaning who takes which feature and we can decide to inject our own work into the schedule if it's something we need to do or will help us), all of my co-workers including my direct boss are pretty great people, we can work from home when we feel like it, and the office is an environment that's generally conducive to getting poo poo done. That's... sort of the minimum for a good tech company, not outstanding. I expect all of these things, I don't get delighted by them. Maybe I'm just spoiled at this point?
|
|
#
?
Mar 12, 2017 12:40
|
|
|
Volmarias posted:That's... sort of the minimum for a good tech company, not outstanding. I expect all of these things, I don't get delighted by them. Lot of bad tech companies out there though, which is why I have been content (don't believe I used language that implied I was delighted with them) to stay where I am. But again, dark clouds on the horizon.
|
|
#
?
Mar 12, 2017 12:55
|
|
|
Che Delilas posted:Lot of bad tech companies out there though, which is why I have been content (don't believe I used language that implied I was delighted with them) to stay where I am. But again, dark clouds on the horizon. Yes, but you can also be content at a place that won't underpay you.
|
|
#
?
Mar 12, 2017 16:46
|
|
|
Che Delilas posted:Lot of bad tech companies out there though
|
|
#
?
Mar 13, 2017 00:02
|
|
|
Last summer I picked up an internship with a pretty large raw material manufacturing company in their R&D department. I worked as an intern under one of their senior EE's who is more or less their senior systems guy. The whole R&D group is comprised of engineers, a few PhD chemists, and engineering technicians. The guy I worked for and his colleague make up what is essentially their systems automation/data analysis/DBA/etc... shop that supports the efforts of the rest of the R&D staff as well as automation in the production environment. I ended up getting a full time offer that I accepted on at the beginning of the year. The compensation is great for my area, and I feel like I'm a good fit for the culture at the company. With that being said, I'm the first full time CS major they've hired and, according to the executive of our department, will probably not be the last. This essentially puts me in a position of being able to sort of build our development shop's operations (obviously with some oversight from my boss) and the prospect is a little overwhelming. As it stands the two guys that I worked with during my internship have for the last 10 years or so managed all of their own servers/DBs/code by just sort of picking the simplest method and sticking with it. No version control, no differentiation between production and development environments. Typically their problems have been solved by justifying new hardware/software purchases through the department budged and just throwing money at it until it works. I'd like to be able to come in with a loose framework of a plan to at least develop a system that I can work under and once we start hiring more CS people (they've already been given approval to hire another summer intern) expand upon. The primary things I know I'll be doing on a regular basis are data warehousing, web development for our data visualization platform, and C#/.NET application development for the labs. On top of that, during the interview process our department executive wanted to stress that he is very interested in incorporating data mining/ML (he was heavy on the buzz words) into their processes. They recently spent a good deal on acquiring a SAS license. This industry, and this company in particular, moves slow so I'm not stressed out about trying to wrangle all of this at the same time or within the first year. Likewise, I was told during the interview (and observed first hand during my internship) that the company very much treats their R&D staff as independent agents who are generally responsible for initiating their own projects and managing essentially a priority queue of requests from other R&D staff and production engineers. As far as the data mining/ML thing goes I've taken a few stats classes, AI, data mining, and a natural language processing course and feel comfortable with the general concepts. I've been looking around at commonly used python libraries, MATLAB, and a few other options for building predictive models. We have a ton of data, but it is structured pretty poorly at the moment. From what I've read SAS seems to be on the decline, and is probably not actually cost effective at the stage we're at right now (ground zero)? If anyone has any advice on setting up a development shop in an R&D environment, or generally working in an environment like this I'd greatly appreciate it.
|
|
#
?
Mar 13, 2017 00:47
|
|
|
Volmarias posted:Yes, but you can also be content at a place that won't underpay you. Vulture Culture posted:Your goal is to figure out how to filter the good companies from the bad ones, not just to give up and point to the numbers game as the reason why you can't have better. Of course I can have better, but as I keep saying there's risk involved that so far I haven't been willing to take, how is that hard to understand?
|
|
#
?
Mar 13, 2017 01:04
|
|
|
Portland Sucks posted:Last summer I picked up an internship with a pretty large raw material manufacturing company in their R&D department. I worked as an intern under one of their senior EE's who is more or less their senior systems guy. The whole R&D group is comprised of engineers, a few PhD chemists, and engineering technicians. The guy I worked for and his colleague make up what is essentially their systems automation/data analysis/DBA/etc... shop that supports the efforts of the rest of the R&D staff as well as automation in the production environment. Do you have actual buy in from management for force changes, such as "no code that's not checked in makes it to production, check in is gated by a build server, code reviews are mandatory" or are you going to be spackling layers of "machine learning" on top of the wild west? Che Delilas posted:Of course I can have better, but as I keep saying there's risk involved that so far I haven't been willing to take, how is that hard to understand? It sounds like you're trying to convince yourself more than us. Of course there's risk, but there's also a risk that you'll continue to stagnate and you'll eventually miss out on six or more figures of potential income that you squandered because you were afraid that you'd go from a shop that's possibly going to turn bad to one that's bad but strangely also pays well and understands that churn is expensive. I'll admit that I did end up with that job briefly, but then I got out after less than a year and became much happier (and better paid) at the next one.
|
|
#
?
Mar 13, 2017 01:53
|
|
|
Volmarias posted:Do you have actual buy in from management for force changes, such as "no code that's not checked in makes it to production, check in is gated by a build server, code reviews are mandatory" During my internship I had an essentially continuous discussion with my boss about this topic. As an anecdote to demonstrate the current feeling within R&D: our IT and business applications side of the house has typical policies like that and early on (over 15 years now) the R&D group started buying and maintaining their own servers because they felt the rigidity of getting approval through an outside department in order to implement changes was not suitable for the operational pace they worked at. So now R&D has zero support from IT as far as their internal servers are concerned, but they have full autonomy over them. In the meantime, as R&D engineers started to develop applications, create databases logging sensor outputs from machinery, etc.. that would be used in someway on the floor they got into the rhythm of just being able to create something and try it out live so there really isn't a distinction between a development and production phase with these tools. One of the things that my boss (he'll cease to be my boss when I start full time) stressed to me throughout the last year is that he's very aware that their development methods are thrown together haphazardly. He is very open to changes given sufficient justification. Previously, his primary concern was keeping their internal systems as simple as possible so that anyone else could potentially pick up where he left off. That thought process was specifically tailored around working with mechanical, chemical, and a few electrical engineers. However, now that they've apparently committed to slowly building up a team of developers I feel like I'll have sufficient justification to argue for the implementation of some SE practices within the department. Even if its just me doing it at first if I can prove that its compatible with their operations I think that is where I would find the buy in. quote:or are you going to be spackling layers of "machine learning" on top of the wild west? This is going to be a tangential effort that will have to be a slow burn I think. During the interview w/ our exec he used the terms "data mining, deep learning, neural network, AI" freely and interchangeably and really really wanted to emphasize that he was convinced that it was ~~the future~~. I haven't had an opportunity to talk with him again since the interview so I'm running on minimal information in regards to what he actually wants to happen and how, but at the moment all I know is that he is apparently intensely excited about buzzwords concerning ML, approved spending six figures on a SAS license that hasn't been touched in over six months, and also hired an undergrad CS major to being this project. My thoughts so far are just that I need to sit down with him again as well as some of the other staff and start a discussion about what areas of the operations that we collect data on should be within the scope, spend some more time becoming familiar with the domain, and then just begin with some cursory data exploration to get the ball rolling while implementing some sort of warehousing plan.
|
|
#
?
Mar 13, 2017 03:22
|
|
|
You talk a lot but it can be distilled as: "no, not really" and "unfortunately yes". Please keep posting for our amusement but keep in mind we warned you. My predictions ate that you will spend a few months getting approval for a git or other vcs server, a month or two explaining the workings to people, some weeks or months where it is used badly and then the moment comes where there is a production incident, all qa gates are overruled, the vcs is seen as cumbersome and is abandoned and about a year of work you did is pushed aside. A few weeks later, when you are browsing the Internet for new roles and find that nobody want a dev with irrelevant experience, the hotfix that caused the downgall of your efforts will blow up yhe production system in a spectacular way for which you will be blaimed as your qa system is bad and you should feel bad. You are let go with fault for gross underperformance and are on the street with no unemployment or savings. Moral of the story: Better be careful in wanting to improve homebrew setups build by non-devs because you will be sucking dicks for money within yhe year.
|
|
#
?
Mar 13, 2017 07:45
|
|
|
Holy gently caress dude. Reloading the page is not a valid fix for getting the data in your stupid widget to update in our React SPA. Just follow the god damned pattern we use everywhere else and it'll magically work. Him: merges it anyway.
|
|
#
?
Mar 13, 2017 14:49
|
|
|
Gounads posted:Holy gently caress dude. Reloading the page is not a valid fix for getting the data in your stupid widget to update in our React SPA. Just follow the god damned pattern we use everywhere else and it'll magically work. poo poo like that is why I'm really glad we have a super-strict code review process. If your code is poo poo it ain't making it to master.
|
|
#
?
Mar 13, 2017 14:59
|
|
|
|
| # ? May 12, 2024 02:11 |
|
|
Clanpot Shake posted:poo poo like that is why I'm really glad we have a super-strict code review process. If your code is poo poo it ain't making it to master. We're supposed to. But I'm just contractor dude and he's long-term guy.
|
|
#
?
Mar 13, 2017 15:20
|
|