kk
Lables are 2*(samples/40)
Size: a a a
2018 July 06
YB
Looks like a very good dataset!
kk
Would you please make me a little clear about timestep definition here in my case
YB
So, basically, you could read DeepSpeech2 for an idea of what architecture you can take
YB
You can group them, yes, with some conv layer on the front
YB
Or can pass them individually
YB
Or you can pass them in parallel but without a conv layer, but for a LSTM it'll be a little be harder to figure out all connections in the data
kk
I already used the conv1d with lstm and even individually lstm but im stocked in very basic question. Here that what is the timestep here for me
YB
For most packages you need to flatten both dimensions into one then
YB
But other than this, feels right
YB
Like, (T*F, batchsize)
kk
For most packages you need to flatten both dimensions into one then
Let say i have only one lable to not need to flatten it
kk
Like, (T*F, batchsize)
So it mean (timestep*featuresize,batchsize)? Right?
YB
Also, if you would pass the data individually and use a custom metric with checks every 40 steps, it won't have any benefit over this grouped solution, I guess.
kk
So it mean (timestep*featuresize,batchsize)? Right?
So my question is what is timestep here should i set it 40
kk
Also, if you would pass the data individually and use a custom metric with checks every 40 steps, it won't have any benefit over this grouped solution, I guess.
What do you mean exactly it's not clear to me
YB
So my question is what is timestep here should i set it 40
Yes. Packs of 40 timesteps loaded at once should work the best
kk
So then what do you suggest for intera person policy? It seems that model cannot find common pattern through different people
kk
Someone suggested me to split everyone's sinal in 10second for instand and shuffel thi bucket of 10sec then feed the network with a larg batches