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/
ImQ009 has joined #mlpack
ImQ009 has quit [Quit: Leaving]
ImQ009 has joined #mlpack
favre49 has joined #mlpack
favre49 has quit [Ping timeout: 260 seconds]
favre49 has joined #mlpack
< zoq>
nishantkr18[m]: Btw. it might be faster to train the model on your own machine first, not sure.
favre49 has quit [Remote host closed the connection]
kwanikar7 has joined #mlpack
kwanikar7 has quit [Remote host closed the connection]
< nishantkr18[m]>
zoq: Yeah, I do all tests on my machine :)
< zoq>
nishantkr18[m]: :)
< zoq>
nishantkr18[m]: I tested 25 but the agent was already pretty good, let me rerun the test
< nishantkr18[m]>
zoq: Hmm.. sometimes its able to get 200 on 100 consec trials, with threshold kept at 20!
< zoq>
nishantkr18[m]: yes, I think this is what I see here
< nishantkr18[m]>
zoq: although that happens with a larger network
< nishantkr18[m]>
zoq: I think we might want to decrease the network size, just to make sure it dosent train so fast :-D
< zoq>
nishantkr18[m]: yes, I mean the output is nice, but since we like to show the agent at different stages, you can't really see the point
< nishantkr18[m]>
<zoq "nishantkr18: https://gym.kurg.or"> zoq: Actually I think this happens with a random agent at play..
< zoq>
and that's a good example for the second stage
< zoq>
nishantkr18[m]: We can keep the settings if you like.
< nishantkr18[m]>
<zoq "and that's a good example for th"> Yup, I guess we get that at the end of the second training right?
< zoq>
yes
< zoq>
so everything looks good
< nishantkr18[m]>
<zoq "nishantkr18: We can keep the set"> I'm not sure either.. Sometimes the intermediate threshold is so bad, and at other times, it solves it the problem entirely..
< zoq>
nishantkr18[m]: for the first first stage right?
< nishantkr18[m]>
<zoq "nishantkr18: for the first first"> Yes
< zoq>
nishantkr18[m]: We could go for number of episodes instead of avg reward.
< zoq>
just break after the first iteration :)
< nishantkr18[m]>
zoq: Yeah.. I hope that works..
< zoq>
nishantkr18[m]: It's a good 'problem' to have.
< nishantkr18[m]>
zoq: Shall I try that?
< nishantkr18[m]>
<zoq "nishantkr18: It's a good 'proble"> 😆
< zoq>
nishantkr18[m]: Sure, if that dosn't work we can keep it.
< nishantkr18[m]>
Ok :)
< zoq>
nishantkr18[m]: Good 'problem' - that the agent is already quite good after a couple of episodes
< nishantkr18[m]>
😁
< nishantkr18[m]>
zoq: With a 110 episodes training, it get to 13 on 100consec trials.
< nishantkr18[m]>
zoq: Let me increase it a bit and see
< zoq>
nishantkr18[m]: Sounds good.
jeffin143 has joined #mlpack
< jeffin143>
zoq[m] , rcurtin : any of you still up ?
< nishantkr18[m]>
zoq: and with 120 episodes, it reaches 167
ImQ009 has quit [Quit: Leaving]
jeffin143 has quit [Remote host closed the connection]
< rcurtin>
jeffin143: I'm here now... but I guess you just left :(
< HimanshuPathakGi>
Hey @saksham189 My rbfn is giving an classification error 0.30400 but when I am compiling it by creating a speperate file by copying same test code I am getting classification error of 0.228 I set the threshold to 0.31 also I want to ask about activation functions should I add them by adding separate file for each or by adding them inside rbfn code. What do you suggest??