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/
< walragatver[m]>
birm: Hey birm, I am having a small meet with jeffin in next 40 minutes. Sorry, I forgot to update you about that. Would you be able to attend it?
< walragatver[m]>
And yes it's an IRC meet
< kartikdutt18Gitt>
Great. Thanks a lot.
< birm[m]>
@walragatver:matrix.org: on this channel?
< walragatver[m]>
Yeah
< jeffin143[m]>
:)
< walragatver[m]>
Hey jeffin let's start
< jeffin143[m]>
Sure
< walragatver[m]>
birm we are starting the meet
< RyanBirminghamGi>
ok!
jeffin143 has joined #mlpack
< walragatver[m]>
So jeffin, anything you would like to say?
< walragatver[m]>
Any updates from your side?
< jeffin143>
yes I did go through the scaler class but I has some doubts
< jeffin143>
I can understand the file write side of summary class but when it doesn't use file write where does the summary gets written ?
< jeffin143>
in memory ?
< walragatver[m]>
Yeah actually I also went through it
< jeffin143>
Now Sometime it is really confusing , because they are using two summary class one from tensorflow and one from tensorboard , here are the two files
< walragatver[m]>
Yeah, I have gone through them. Basically, the problem is some of the functions are generated during build in tensorflow so we cannot view them.
< walragatver[m]>
like gen_summary() function
< walragatver[m]>
gen part signifies it's generated during file.
< jeffin143>
Ohh I see, I haven't be accustomed to tensor flow , I did go through extensively when I wanted to view the scaler functionality
< walragatver[m]>
But yeah I hope you got an idea how their api is implemented basically the design
< walragatver[m]>
There is one filewriter object with which they all various summaries including scalars, images etc.
< jeffin143>
Yes , but I am not sure how do they redirect everything to specific folder ?
< walragatver[m]>
log*
< walragatver[m]>
There is one event file for a particular tensorflow session() generally a session can be considered as a single program / model.
< walragatver[m]>
So I think we should also do some stuff in the similar way, what do you think?
< walragatver[m]>
<jeffin143 "Yes , but I am not sure how do t"> Yeah I will comeback to it soon.
< jeffin143>
Ok Now your mail required me to make a table of the things we log
< jeffin143>
So Now I wanted to know, Should we log the accuracy of a linear regression model ?
< jeffin143>
Because I am not sure what would be the use case for that. ?
< walragatver[m]>
Okay, Have you made the list? If yes, just mail me out all things which you think we should log. I will look at them soon.
< walragatver[m]>
<walragatver[m] "So I think we should also do som"> What's your view on this?
< jeffin143>
walragatver[m]: The API I discussed was similar to that, An Event or session was project id with time stamp similar to what tensorflow does but the issue was the subdirecotry one `train/` `validation` and `metrics`
< jeffin143>
with same file write object
< jeffin143>
fileWriter object *
< RyanBirminghamGi>
I should get to know the tf source better so I could have more insight here....
< walragatver[m]>
Okay yeah. I have written some stuff about it. We will use a single file. I have noted advantages using single file in a doc. I will share it with you at the end of this meeting. Basically there are some more things I need to discuss today. So first lets discuss them.
< walragatver[m]>
birm: It's fine. No problems.
< RyanBirminghamGi>
I'll find some time to explore it :)
< jeffin143>
Yes , Now < Okay, Have you made the list? > Basically this is all the ann models which can call callbacks ? Right ?
< jeffin143>
Train Accuracy graph against epoch Validation Accuracy graph against epochTrain Loss graph against epochValidation graph against epoch
< walragatver[m]>
jeffin143: birm: This is just the first meet. So it's fine.
< walragatver[m]>
<jeffin143 "Yes , Now < Okay, Have you made "> Yeah I think callbacks would be present whenever we have used ensmallen callbacks.
< jeffin143>
Ok walragatver[m], So I want to make some things sure before we move ahead
< walragatver[m]>
optimizers I mean to say.
< walragatver[m]>
Okay so let's shift to the main agenda now.
< jeffin143>
We are going to use a summary class -> Which will log some summary to files , And then We have to come up with.a callback that will use this summary to log different information , Also i use want he can use this summary class to log more information , apart from the default ones we have provided
< jeffin143>
if user*
< walragatver[m]>
jeffin143: birm: I think it's important. that we have a meeting twice in a week. What do you think?
< jeffin143>
Definitely, I am lost
< jeffin143>
I am in for twice a week meeting
< RyanBirminghamGi>
Sounds reasonable!
< RyanBirminghamGi>
I keep missing IRC messages, so I need to figure out why that is...
< walragatver[m]>
Let's keep one on weekday and one on weakend. Let me know what you think. So, I think one on sunday and one on wednesday or thrusday what do you say?
< walragatver[m]>
<RyanBirminghamGi "I keep missing IRC messages, so "> Yeah it's fine. No problems.
< jeffin143>
sunday and Wednesday looks alright to me, I am ok with these days
< RyanBirminghamGi>
If you can, just send an email if it looks like I've missed something, just in case :)
< walragatver[m]>
birm: sure I will do that.
< RyanBirminghamGi>
Thank you!
< RyanBirminghamGi>
Sun and wed are also especially good for me
< walragatver[m]>
Next thing. jeffin143: I hope you are also in final year?
< jeffin143>
yes
< walragatver[m]>
Fine so Sunday and Wednesday it is.
< jeffin143>
Yes cool :+1
< walragatver[m]>
Do you have any other commitments currently?
< walragatver[m]>
<walragatver[m] "Do you have any other commitment"> This was for jeffin143
< jeffin143>
No, Currently there are no calls or joining dates information from my org where I have been placed due to corona issue
< walragatver[m]>
Okay great.
< jeffin143>
yes
< walragatver[m]>
jeffin143: birm: So, next important question is that do you think that it would be feasible to implement both backend and frontend of the dashboard in the GSOC period?
< RyanBirminghamGi>
As far as I'm aware, the backend is more essential. For the frontend, I'd start with a minimal proof of concept just to demonstrate that it works.
< jeffin143>
Yes , I think both would be feasible If we are concrete on the API Side, I mean the Logging Part, I am sure it wouldn't be much difficult implementing it , the design phase is diffiicult
< walragatver[m]>
jeffin143: birm: Yes, both of you are correct. Basically there is also an alternative which is to integrate with tensorboard. What's your view on that?
< jeffin143>
I am not sure how would we do that, I tried taking a look at that, the format which tensor-flow uses cannot be read so I was not sure how to log in that format, but if we are able to generate the same log structure as in tensorflow
< walragatver[m]>
Basically there are some pros and cons of both the things.
< walragatver[m]>
I will start with the our own dashboard
< RyanBirminghamGi>
jeffin143: do you think it'd be faster/easier to do a minimal proof of concept, or integrate with tensorboard?
< jeffin143>
@Rya
< jeffin143>
RyanBirminghamGi : I guess it would be really powerful/faster if we redirect our metrics to tensor board but I am not sure how would we achieve it
< walragatver[m]>
<jeffin143 "I am not sure how would we do th"> Yeah, they use protocol buffers. I have dome some research on it. I will share about it as well.
< jeffin143>
walragatver[m] : If we can do that, I think so we can solely devote our attention to backend and it would be really good, tensorboard is powerful
< jeffin143>
I am not sure about the con you are talking about it ?
< walragatver[m]>
jeffin143: birm: Currently let's keep our backend design such that we can adopt in any of the alternatives in future. What do you guys say?
< RyanBirminghamGi>
My perspective on these kinds of vis related projects has been that it could be more useful to start with minimal versions of each component, so that we have a working, if bad, version sooner.
< jeffin143>
walragatver[m] : I agree with you , so our backend should be as similar to what tensorflow has so that if we are opting for tensorboard then we can use that or use our custom one
< walragatver[m]>
birm: Yes correct. I think we will just focus on plotting scalars firstly. They are quit easy to work upon. What do you think?
< walragatver[m]>
These are some attempts in c++ backend for standalone tensorboard.
< walragatver[m]>
So we can refer them
< jeffin143>
Ok I will take a look at them
< walragatver[m]>
jeffin143: birm: Also go through these implementations in your free time.
< jeffin143>
walragatver[m] : Should we log in raw form or use proto buffers ?
< walragatver[m]>
<jeffin143 "walragatver : Should we log in r"> I think proto buffers would be good an faster to implement as they use XML. I am not sure about it I think birm may help us with this.
< walragatver[m]>
So this is all from my side. Sorry, I was talking too much in this meet.
< RyanBirminghamGi>
You're the main mentor, so no surprise :)
< RyanBirminghamGi>
No issue with protobufs, as it should be easier to work with.
< walragatver[m]>
birm: jeffin143: Anything else you want to discuss?
< jeffin143>
walragatver[m] : You said you would be sharing a docs ?
< jeffin143>
or was it the repo
< jeffin143>
?
< walragatver[m]>
jeffin143: Oh yes. Basically I was also working on the POC but then encountered with the repo. Just a sec I will share it with you.
< walragatver[m]>
jeffin143: birm: This was just an attempt from my side. Might be wrong for sure. Feel free to leave your thoughts on it. You can comment as on GSOC proposal.
< jeffin143>
Ok , I will take a look , Thanks for all the hard work :)
< walragatver[m]>
No worries. With this we can conclude our meet.
< walragatver[m]>
birm: jeffin143: Thank you for your time.
< jeffin143>
Sure , Let's meet on wednesday , Good night
< walragatver[m]>
jeffin143: feel free to ask any doubts you have. I would keep looking the irc.
< birm[m]>
What time on Wednesday do we have planned?
< jeffin143[m]>
Same time , works for me @brim : if you want you can suggest an alternative
< RyanBirminghamGi>
That should be fine, adding to my calendar
jeffin143 has quit [Remote host closed the connection]
ImQ009 has quit [Read error: Connection reset by peer]
< HimanshuPathakGi>
Hey @saksham189 I think I implemented the RBFN code correctly but it is still failing with mnist_first_250_training_4s_and_9s.arm
M_slack_mlpack_4 has joined #mlpack
M_slack_mlpack_4 is now known as jjb[m]
< jjb[m]>
Howdy all from #mlpack:matrix.org slack 🙂
< zoq>
jjb[m]: Hello there!
< jjb[m]>
Bit surreal seeing messages replicated out in realtime to many different platforms (IRC, Slack, Gitter) .