State of the project?

[ghost]

Hi Simon,

I was just wondering about the state of the project? Was everything running, I mean the examples nuk, coco and shape or is it still in alpha phase?
I just want to know how much time I would have to put in in order to make it run?
thx and
Regards

[simonm3]

Everything was running including the examples. Easiest would be to use pytorch v1.0 but find the source code for the deprecated package; and copy it into your pythonpath. I recall it was just one file. Then everything should work. May also work with more recent pytorch but not tried. Was mostly for learning and I have not used since. One learning is that models like this are really hard to debug and test. If you make some changes then make small changes and test often. If you make 100 changes and it no longer gives good output then tricky to pinpoint the error.

…

On Fri, 27 Dec 2019 at 20:32, 1seck ***@***.***> wrote: Hi Simon, I was just wondering about the state of the project? Was everything running, I mean the examples nuk, coco and shape or is it still in alpha phase? I just want to know how much time I would have to put in in order to make it run? thx and Regards — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2?email_source=notifications&email_token=AAJE32LW5IB2WK5KJ63ZDPDQ2ZQ4FA5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IC5IFAQ>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJE32IOIJLELRWAM5YEUHTQ2ZQ4FANCNFSM4KABUXPQ> .

[ghost]

Hi Simon, answer inline

Am 28.12.2019 um 16:15 schrieb simon ***@***.***>: Everything was running including the examples. Easiest would be to use pytorch v1.0 but find the source code for the deprecated package; and copy it into your pythonpath. I recall it was just one file. Then everything should work. May also work with more recent pytorch but not tried.

just one file, ok. so pytorch 1.0 is also deprecated and downgrading to 0.4 seems too big of a step back. there are help github-issues on it but 2 packages have to be replaced and also the variable types of the function call args are different (tensor vs numpy). or do you mean to make ffi work instead of changing to cpp_extension?

Was mostly for learning and I have not used since. One learning is that models like this are really hard to debug and test. If you make some changes then make small changes and test often. If you make 100 changes and it no longer gives good output then tricky to pinpoint the error.

thx, sounds like what everyone is saying about ml, dl. sorry to ask again, but you did have comparable results? despite of that thx again :)

…

On Fri, 27 Dec 2019 at 20:32, 1seck ***@***.***> wrote: > Hi Simon, > > I was just wondering about the state of the project? Was everything > running, I mean the examples nuk, coco and shape or is it still in alpha > phase? > I just want to know how much time I would have to put in in order to make > it run? > thx and > Regards > > — > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub > <#2?email_source=notifications&email_token=AAJE32LW5IB2WK5KJ63ZDPDQ2ZQ4FA5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IC5IFAQ>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAJE32IOIJLELRWAM5YEUHTQ2ZQ4FANCNFSM4KABUXPQ> > . > — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#2?email_source=notifications&email_token=AB6B2R3Y4TUEQKPY3KXEFELQ25UQ3A5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHYLZ6A#issuecomment-569425144>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB6B2RZKNBGG7XZ5TYK745TQ25UQ3ANCNFSM4KABUXPQ>.

[simonm3]

Easiest is get it running with V1 + copy the cffi file from older version.....then you dont need to upgrade to cpp. It was all running with that. Once that working then you could upgrade to V2 - no idea what is required for that. Yes results with V1 were almost identical to keras version. The issue is every change you make is hard to test. You can see from diagram it is complex algorithm. Especially the ranking of regions means a tiny change can change the selected regions so you end up with different results.

…

On Sat, 28 Dec 2019 at 17:52, 1seck ***@***.***> wrote: Hi Simon, answer inline > Am 28.12.2019 um 16:15 schrieb simon ***@***.***>: > > Everything was running including the examples. Easiest would be to use > pytorch v1.0 but find the source code for the deprecated package; and copy > it into your pythonpath. I recall it was just one file. Then everything > should work. May also work with more recent pytorch but not tried. > just one file, ok. so pytorch 1.0 is also deprecated and downgrading to 0.4 seems too big of a step back. there are help github-issues on it but 2 packages have to be replaced and also the variable types of the function call args are different (tensor vs numpy). or do you mean to make ffi work instead of changing to cpp_extension? > Was mostly for learning and I have not used since. One learning is that > models like this are really hard to debug and test. If you make some > changes then make small changes and test often. If you make 100 changes and > it no longer gives good output then tricky to pinpoint the error. > thx, sounds like what everyone is saying about ml, dl. sorry to ask again, but you did have comparable results? despite of that thx again :) > On Fri, 27 Dec 2019 at 20:32, 1seck ***@***.***> wrote: > > > Hi Simon, > > > > I was just wondering about the state of the project? Was everything > > running, I mean the examples nuk, coco and shape or is it still in alpha > > phase? > > I just want to know how much time I would have to put in in order to make > > it run? > > thx and > > Regards > > > > — > > You are receiving this because you are subscribed to this thread. > > Reply to this email directly, view it on GitHub > > < #2?email_source=notifications&email_token=AAJE32LW5IB2WK5KJ63ZDPDQ2ZQ4FA5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IC5IFAQ >, > > or unsubscribe > > < https://github.com/notifications/unsubscribe-auth/AAJE32IOIJLELRWAM5YEUHTQ2ZQ4FANCNFSM4KABUXPQ > > > . > > > — > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub < #2?email_source=notifications&email_token=AB6B2R3Y4TUEQKPY3KXEFELQ25UQ3A5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHYLZ6A#issuecomment-569425144>, or unsubscribe < https://github.com/notifications/unsubscribe-auth/AB6B2RZKNBGG7XZ5TYK745TQ25UQ3ANCNFSM4KABUXPQ >. > — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2?email_source=notifications&email_token=AAJE32INICRFPDJRQ2MKRXDQ26G65A5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHYO7KY#issuecomment-569438123>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJE32M4PQAXFXCXRYVZTKTQ26G65ANCNFSM4KABUXPQ> .

[ghost]

Good to hear that. Of course it's lot of parameters and hyper parameters... And probably slightly different model designs. But also a robust algorithm So I will try to copy the ffi file. Almost had the cpp running but still thx. That's deep learning like Jeremy said more difficult to debug. I will try my best then. Regards simon <notifications@github.com> schrieb am Sa., 28. Dez. 2019, 19:29:

…

Easiest is get it running with V1 + copy the cffi file from older version.....then you dont need to upgrade to cpp. It was all running with that. Once that working then you could upgrade to V2 - no idea what is required for that. Yes results with V1 were almost identical to keras version. The issue is every change you make is hard to test. You can see from diagram it is complex algorithm. Especially the ranking of regions means a tiny change can change the selected regions so you end up with different results. On Sat, 28 Dec 2019 at 17:52, 1seck ***@***.***> wrote: > Hi Simon, > > answer inline > > > Am 28.12.2019 um 16:15 schrieb simon ***@***.***>: > > > > Everything was running including the examples. Easiest would be to use > > pytorch v1.0 but find the source code for the deprecated package; and > copy > > it into your pythonpath. I recall it was just one file. Then everything > > should work. May also work with more recent pytorch but not tried. > > > just one file, ok. > so pytorch 1.0 is also deprecated and downgrading to 0.4 seems too big of > a step back. there are help github-issues on it but 2 packages have to be > replaced and also the variable types of the function call args are > different (tensor vs numpy). > or do you mean to make ffi work instead of changing to cpp_extension? > > Was mostly for learning and I have not used since. One learning is that > > models like this are really hard to debug and test. If you make some > > changes then make small changes and test often. If you make 100 changes > and > > it no longer gives good output then tricky to pinpoint the error. > > > thx, sounds like what everyone is saying about ml, dl. > sorry to ask again, but you did have comparable results? > > despite of that thx again :) > > > > On Fri, 27 Dec 2019 at 20:32, 1seck ***@***.***> wrote: > > > > > Hi Simon, > > > > > > I was just wondering about the state of the project? Was everything > > > running, I mean the examples nuk, coco and shape or is it still in > alpha > > > phase? > > > I just want to know how much time I would have to put in in order to > make > > > it run? > > > thx and > > > Regards > > > > > > — > > > You are receiving this because you are subscribed to this thread. > > > Reply to this email directly, view it on GitHub > > > < > #2?email_source=notifications&email_token=AAJE32LW5IB2WK5KJ63ZDPDQ2ZQ4FA5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IC5IFAQ > >, > > > or unsubscribe > > > < > https://github.com/notifications/unsubscribe-auth/AAJE32IOIJLELRWAM5YEUHTQ2ZQ4FANCNFSM4KABUXPQ > > > > > . > > > > > — > > You are receiving this because you authored the thread. > > Reply to this email directly, view it on GitHub < > #2?email_source=notifications&email_token=AB6B2R3Y4TUEQKPY3KXEFELQ25UQ3A5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHYLZ6A#issuecomment-569425144 >, > or unsubscribe < > https://github.com/notifications/unsubscribe-auth/AB6B2RZKNBGG7XZ5TYK745TQ25UQ3ANCNFSM4KABUXPQ > >. > > > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > < #2?email_source=notifications&email_token=AAJE32INICRFPDJRQ2MKRXDQ26G65A5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHYO7KY#issuecomment-569438123 >, > or unsubscribe > < https://github.com/notifications/unsubscribe-auth/AAJE32M4PQAXFXCXRYVZTKTQ26G65ANCNFSM4KABUXPQ > > . > — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#2?email_source=notifications&email_token=AB6B2R4YR5ON5TDVIAIYZ3DQ26LHBA5CNFSM4KABUXP2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHYPT5I#issuecomment-569440757>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB6B2RZR4EHPMNFMBI6FUBTQ26LHBANCNFSM4KABUXPQ> .