# machine-learning
- h
hecko
04/07/2022, 9:51 AMbecause the regular only does 25% arpabet for some reason - h
hecko
04/07/2022, 9:51 AMmine doesn't quite do 100% because of a bug but i can fix that when i get back home - g
Gosmokeless28
04/07/2022, 9:52 AMIn that case, should I replace the one that's in #841437191073955920? - h
hecko
04/07/2022, 9:52 AMmine is a bit of a mess in other ways lemme just give you the change that matters - h
hecko
04/07/2022, 9:53 AM-or well not yet because ahem when i get back home i'll fix it up so it's 100% arpabet and then i'll give you the patch - g
Gosmokeless28
04/07/2022, 9:53 AMWdym give me the patch? - h
hecko
04/07/2022, 9:54 AMthere's one line of code that needs to be changed - h
hecko
04/07/2022, 9:54 AMspecifically in the setup step at the bottom there's a thing that downloads the arpabet patch and i'll give you a command that downloads my version of the patch instead - g
Gosmokeless28
04/07/2022, 9:56 AMWhy not put the command in a cell of its own in your notebook? - h
hecko
04/07/2022, 9:57 AMhm - h
hecko
04/07/2022, 9:57 AMmy notebook also has other changes like i said earlier - h
hecko
04/07/2022, 9:57 AMbut y'know what sure i'll include those too - m
mildrefusal
04/07/2022, 5:55 PMHey all, I'm training my first model in colab atm, the loss seems to be dropping nicely but I'm not sure how to interpret the tensorboard feedback graphs. Some epochs return a nice diagonal line, others don't. Does anyone understand why this might be happening and if it suggests that something has gone wrong? Using the standard training Colab (unfortunately it's only giving me a K80) https://colab.research.google.com/drive/1WTilMdm9Vf7KE79gzkeeTBigAN6iv3Bg?usp=sharing#scrollTo=iQBUZTuMogui Using this dataset https://discord.com/channels/768215836665446480/835647732453605376/942467660610936884 - u
{K EY1} (Kei)
04/07/2022, 6:16 PMIt might mean that some samples aren't good quality Do you have some samples that are whispery, noisy, or screamy? - u
(Dawn) Will Draw Fictional Women
04/07/2022, 6:27 PMthe small one just means a short wav - u
(Dawn) Will Draw Fictional Women
04/07/2022, 6:27 PMif the graphs are a mostly solid orange staircase then theyre fine no matter when they end or if something happens afterwards - m
mildrefusal
04/07/2022, 7:59 PMThanks, that’s really helpful! - s
ShawnVA
04/08/2022, 2:25 AMhi im training a model and all the epochs look like this is this a good sign - u
(Dawn) Will Draw Fictional Women
04/08/2022, 2:40 AMthis concerns me - s
ShawnVA
04/08/2022, 2:41 AMwhy - u
(Dawn) Will Draw Fictional Women
04/08/2022, 2:41 AMtheres a Split in the StairCase™️ - s
ShawnVA
04/08/2022, 2:42 AMheres what it sounds like so far its based of MTF from scp containment breach - u
(Dawn) Will Draw Fictional Women
04/08/2022, 2:42 AMand this whole thing is just Goop - u
(Dawn) Will Draw Fictional Women
04/08/2022, 2:42 AMis that straight from the dataset - s
ShawnVA
04/08/2022, 2:43 AMdoes that mean the raw version - s
ShawnVA
04/08/2022, 2:43 AMi edited it with radio sounds - u
(Dawn) Will Draw Fictional Women
04/08/2022, 2:43 AMlike - u
(Dawn) Will Draw Fictional Women
04/08/2022, 2:44 AMis that how it is in the dataset - s
ShawnVA
04/08/2022, 2:44 AMoh no its not - u
(Dawn) Will Draw Fictional Women
04/08/2022, 2:44 AMoPhew