ChanServ changed the topic of #mlpack to: "mlpack: a fast, flexible machine learning library :: We don't always respond instantly, but we will respond; please be patient :: Logs at http://www.mlpack.org/irc/
KimSangYeon-DGU has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong_ has joined #mlpack
xiaohong has quit [Ping timeout: 246 seconds]
xiaohong_ has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
< favre49>
rcurtin Thanks for the reminder, I would have missed the meeting '=D
< rcurtin>
:)
jeffin143 has quit [Read error: Connection reset by peer]
jeffin143 has joined #mlpack
favre49 has quit [Remote host closed the connection]
sreenik_windows has joined #mlpack
jeffin143 has quit [Ping timeout: 258 seconds]
xiaohong has joined #mlpack
< sreenik_windows>
I am trying with Windows this time, hope my mic works. Sadly I'm having to use a 3g hotspot from my cellphone due to issues with my broadband :(
favre49 has joined #mlpack
favre49 has quit [Remote host closed the connection]
< xiaohong>
I am using 4G hotspot too.
< xiaohong>
I think the network will not stale. lol
< sreenik_windows>
Let's hope so :)
favre49 has joined #mlpack
xiaohong_ has joined #mlpack
xiaohong has quit [Ping timeout: 250 seconds]
< sreenik_windows>
My zoom application is quitting at regular intervals, let's see if I can make it through
sreenik_windows has quit [Quit: Ping timeout (120 seconds)]
vivekp has joined #mlpack
ImQ009 has joined #mlpack
KimSangYeon-DGU has quit [Ping timeout: 260 seconds]
xiaohong_ has quit [Read error: Connection reset by peer]
xiaohong has joined #mlpack
favre49 has quit [Remote host closed the connection]
KimSangYeon-DGU has joined #mlpack
< KimSangYeon-DGU>
Thanks for the meeting.
KimSangYeon-DGU has quit [Remote host closed the connection]
< xiaohong>
Hoping we can organize this kind of meeting more~
< xiaohong>
It is a great way to participate in and know the recent changes in mlpack.
< rcurtin>
xiaohong: agreed, I enjoyed that :)
xiaohong has quit [Remote host closed the connection]
< rcurtin>
at least from my end the issue is always just time; if we can make it more of a group effort, it gets easier to have these more often :)
xiaohong has joined #mlpack
xiaohong has quit [Remote host closed the connection]
< rcurtin>
I see, thanks. it looks to me like most of the changes are reverse-compatible there (mostly just adding more parameters to a constructor)
< rcurtin>
so the question is just whether we consider "Action" to be part of the public API, that a user might use
< rcurtin>
let me know what you think---it seems more internal to me, but like I said, I am not an expert
< zoq>
rcurtin: Right, will have to take a closer look; will let you know later today.
< rcurtin>
sounds good
favre49 has joined #mlpack
< favre49>
rcurtin: Action is part of the public API, since that is what someone would use to create an effect in the environment
jeffin143 has joined #mlpack
< favre49>
You could take a look at the tests, but as an example, in NEAT, what we do is activate the network, and use the output to construct an Action instance. This is used to move the environment forward one step, by applying that Action
< jeffin143>
rcurtin : just saw mail, sorry missed the meeting
< rcurtin>
jeffin143: no worries :)
< jeffin143>
I really wanted to join
< rcurtin>
I'm writing up some notes now, I'll get them sent in a minute
< jeffin143>
Mess timing in Indian Colleged
< jeffin143>
Thanks :)
< favre49>
In general, someone working with any of our RL environments would do so with a basic for loop, where the construct an Action using whatever algorithm they're using and then use Sample() to find the next State
< favre49>
That state would be inputted into the algo, and once again return an action
< favre49>
Correct me if I'm wrong, but I think this would be the basic essential component of any test or program a user would create
< favre49>
So in the case of continuous environments, users would previously use action.action[0] to set the action value, but now they use action.action
< favre49>
If we are particular about not breaking backwards compatibility, I guess we could add another double variable to the struct called actionDbl and revert action to an array
< rcurtin>
favre49: thanks for the explanation
< rcurtin>
I could go either way on it---but like I said I am not familiar with exactly how this code gets used
< favre49>
And then we would mention that the old version is deprecated
< rcurtin>
personally I like to avoid too many major version increments, so maybe the actionDbl idea could be done, but I am not sure how much extra work that would be
< rcurtin>
right, yeah, there are some other postponed changes and removals of things for the 4.0.0 release too, so I guess, either we leave the RL code as-is and release 4.0.0, and then *also* go through and remove other deprecated options
< rcurtin>
or we do the actionDbl idea, then leave some comments pointing out that it is deprecated and will be changed in 4.0.0
< rcurtin>
I don't have a strong opinion either way---what do you think?
< favre49>
It's not really any extra work, just some small changes in a bunch of files, I don't mind doing it. At worst it's just a bit cofusing
< rcurtin>
(also zoq and xiaohong's opinions could be useful here, plus anyone else who has an opinion :))
< favre49>
If we have a lot of deprecated stuff, perhaps it would make more sense to make a major version bump?
< favre49>
I'm not sure what the status of that is, i thin
< favre49>
i remember robertohueso saying something about deprecation but thats about it
< rcurtin>
it's mostly just some old option names for bindings, if I remember right
< rcurtin>
this seems to find everything:
< rcurtin>
`grep -r 'mlpack 4.0.0' src/`
< favre49>
Looks like 12 instances, not sure if that's a lot or a little
< favre49>
Only 2 are not related to bindings, so it doesnt seem like its a lot
< favre49>
purely from an optics point of view, though, we should make a major version bump if we're releasing a bunch of cool new things
< favre49>
But in that case, we might have to wait longer before the version bump
< rcurtin>
yeah, I know what you mean, the only thing is if we continue to wait, even longer goes by without things being released :)
< rcurtin>
the Julia and Go bindings seem like great fits for 4.0.0, but it may be several weeks before they are ready
< favre49>
Would also look better when we make the blog post, kind of looks like we've made a huge revamp :P
< favre49>
Ah several weeks sounds like
< favre49>
far too long
< rcurtin>
yeah, agreed
favre49 has quit [Remote host closed the connection]
jeffin143 has quit [Ping timeout: 260 seconds]
mikhail has joined #mlpack
mikhail is now known as lozhnikov_
favre49 has joined #mlpack
< favre49>
I also wanted to mention, if we are starting blog posts, we should redo the blog page to make it more structured and easy to access
< favre49>
Or perhaps, like tensorflow, we could make the posts on medium, which would help with publicity i suppose
favre49 has quit [Remote host closed the connection]
< rcurtin>
favre49: yeah, agreed, maybe just Medium is the easy way to go here
< rcurtin>
personally speaking I don't think I really have time to do much in the way of blog posts
favre49 has joined #mlpack
< favre49>
rcurtin I have no issues writing the posts (and running them by the community first of course)
vivekp has quit [Ping timeout: 272 seconds]
< rcurtin>
favre49: that would be awesome!
< sreenik[m]>
zoq: I have pushed the PR for layer_names.hpp (#1987), let me know if any change is required
lozhnikov_ has quit [Quit: lozhnikov_]
lozhnikov_ has joined #mlpack
favre49 has quit [Remote host closed the connection]
mikhail has joined #mlpack
< sreenik[m]>
Could you also review the PR #1985 (getter/accessor methods for layers) when you have some time and let me know about the changes that are required? It would be good to get that merged in the next release because the mlpack-Torch model converter depends on that.
lozhnikov_ has quit [Ping timeout: 244 seconds]
< sreenik[m]>
Atharva and I have had a discussion in the PR itself and I think not much change is required
lozhnikov has joined #mlpack
mikhail has quit [Quit: mikhail]
< zoq>
sreenik[m]: Great will take a look at each.
< zoq>
What about we revert back to double action[1]; and adapt the open PR's instead, it's probably the easiest solution.
< zoq>
If we can agree on that, I can open another PR.
< ShikharJ>
rcurtin: Hey Ryan, sorry I couldn't join for the meeting. Apparently I'm still a bit jet-lagged.
< rcurtin>
ShikharJ: no worries! hope the flight went okay :)
< rcurtin>
ShikharJ: there were a couple PRs I had questions about for the release; if you want to take a look at the notes email I mentioned them there, and if you can let me know if you think they're likely to be merged in the next couple of days, then I can wait for a release for that
< ShikharJ>
rcurtin: Okay, I'll take a look when I get some time.
< rcurtin>
ShikharJ: sounds good, thanks!
< rcurtin>
ok, I think I found the issue with the gcc build docker images... but it will take a day or two to rebuild them all