Is it possible to segment the definition of the function to be taken the gradient?

[dlmacedo]

Dear friend,

Is it possible to segment the definition of the function to be taken the gradient?

For example:

if x < SomeValue then f(x) = someThing
if x > SomeValue then f(x) = someOtherThing

Thanks,

David

[alexbw]

Conditionals are allowed in non-optimized mode (which is the default)

On Mon, May 9, 2016 at 5:20 PM David Lopes Macedo notifications@github.com
wrote:

Dear friend,

Is it possible to segment the definition of the function to be taken the
gradient?

For examplo,
if x < SomeValue then f(x) = someThing
if x < SomeValue then f(x) = someThing

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub
#118

[dlmacedo]

Dear friends,

Could someone tells me what is the error bellow:

local SeLU = function(input)
  local output
  if torch.le(input, 0) then
    output = 0
  else
    output = input
  end 
  return output
end

I am using cudnn and getting this error:

/home/dlm/torch/install/bin/luajit: /home/dlm/torch/install/share/lua/5.1/nn/Container.lua:67: 0ms | Step: 0ms
In 3 module of nn.Sequential:
In 4 module of nn.Sequential:
/home/dlm/torch/install/share/lua/5.1/nn/Dropout.lua:20: bad argument #1 to 'resizeAs' (torch.CudaTensor expected, got userdata)
stack traceback:
[C]: in function 'resizeAs'
/home/dlm/torch/install/share/lua/5.1/nn/Dropout.lua:20: in function </home/dlm/torch/install/share/lua/5.1/nn/Dropout.lua:16>
[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function </home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:41>
[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
train.lua:151: in function 'opfunc'
/home/dlm/torch/install/share/lua/5.1/optim/sgd.lua:44: in function 'sgd'
train.lua:160: in function 'train'
train.lua:236: in main chunk
[C]: in function 'dofile'
.../dlm/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670

When I use the following code, it works great:

local SeLU = function(input)
  local output = 0.5 * (torch.abs(input) + input)
  return output
end

[alexbw]

Is input always a torch tensor or scalar? You're returning a scalar, which might cause weirdness.

[alexbw]

Try torch.le(input, 0).value instead of torch.le(input, 0)

[dlmacedo]

I am using cuddn...

Input and output are tensors...

The following code is ok:

local seluFunc = function(input)
  local output, aux, aux2, aux3
  aux = (seluc/selua+1)/2
  aux2 = ((1-aux)*torch.abs( input+selub)+aux*( input+selub))-selub
  aux3 = -((1-aux)*torch.abs(-aux2+selub)+aux*(-aux2+selub))+selub
  output = selua*aux3
  return output
end

But I still don't understand what is wrong with the code bellow:

local seluFunc2 = function(input)
  local output, lessX, equalX, graterX, aux1, aux2, aux3
  lessX = torch.lt(input, -selub)
  aux1 = torch.gt(input, -selub)
  aux2 = torch.lt(input, selub)
  equalX = torch.cmul(aux1, aux2)
  graterX = torch.gt(input, selub)
  aux3 = torch.add(lessX,equalX)
  output = torch.add(aux3,graterX)
  return output
end

Full logs:

In 3 module of nn.Sequential:
In 54 module of nn.Sequential:
In 3 module of nn.Sequential:
...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:77: attempt to index local 'input' (a nil value)
stack traceback:
...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:77: in function 'checkInputDim'
...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:130: in function <...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:128>
[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function </home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:78>
[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function </home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:78>
[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function 'backward'
train.lua:152: in function 'opfunc'
/home/dlm/torch/install/share/lua/5.1/optim/sgd.lua:44: in function 'sgd'
train.lua:158: in function 'train'
train.lua:234: in main chunk
[C]: in function 'dofile'
.../dlm/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670

WARNING: If you see a stack trace below, it doesn't point to the place where this error occured. Please use only the one above.
stack traceback:
[C]: in function 'error'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:67: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function 'backward'
train.lua:152: in function 'opfunc'
/home/dlm/torch/install/share/lua/5.1/optim/sgd.lua:44: in function 'sgd'
train.lua:158: in function 'train'
train.lua:234: in main chunk
[C]: in function 'dofile'
.../dlm/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670

[alexbw]

Does it work on plain NN, on the CPU?
On Sun, Jun 19, 2016 at 7:10 AM David Macêdo notifications@github.com
wrote:

I am using cuddn...

Input and output are tensors...

The following code is ok:

local seluFunc = function(input)
local output, aux, aux2, aux3
aux = (seluc/selua+1)/2
aux2 = ((1-aux)torch.abs( input+selub)+aux( input+selub))-selub
aux3 = -((1-aux)torch.abs(-aux2+selub)+aux(-aux2+selub))+selub
output = selua*aux3
return output
end

But I still don't understand what is wrong with the code bellow:

local seluFunc2 = function(input)
local output, lessX, equalX, graterX, aux1, aux2, aux3
lessX = torch.lt(input, -selub)
aux1 = torch.gt(input, -selub)
aux2 = torch.lt(input, selub)
equalX = torch.cmul(aux1, aux2)
graterX = torch.gt(input, selub)
aux3 = torch.add(lessX,equalX)
output = torch.add(aux3,graterX)
return output
end

Full logs:

In 3 module of nn.Sequential:

In 54 module of nn.Sequential:

In 3 module of nn.Sequential:

...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:77: attempt to
index local 'input' (a nil value)
stack traceback:
...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:77: in
function 'checkInputDim'
...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:130: in
function <...lm/torch/install/share/lua/5.1/nn/BatchNormalization.lua:128>

[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function
'rethrowErrors'

/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function

[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function
'rethrowErrors'

/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function

[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function
'rethrowErrors'

/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function
'backward'
train.lua:152: in function 'opfunc'

/home/dlm/torch/install/share/lua/5.1/optim/sgd.lua:44: in function 'sgd'

train.lua:158: in function 'train'
train.lua:234: in main chunk

[C]: in function 'dofile'
.../dlm/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main
chunk
[C]: at 0x00406670

WARNING: If you see a stack trace below, it doesn't point to the place
where this error occured. Please use only the one above.
stack traceback:
[C]: in function 'error'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:67: in function
'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:84: in function
'backward'
train.lua:152: in function 'opfunc'

/home/dlm/torch/install/share/lua/5.1/optim/sgd.lua:44: in function 'sgd'

train.lua:158: in function 'train'
train.lua:234: in main chunk

[C]: in function 'dofile'
.../dlm/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main
chunk
[C]: at 0x00406670

—

You are receiving this because you commented.

Reply to this email directly, view it on GitHub
#118 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/AAJ4j6iVC_i76tfAijynMxiBkjTK-qMyks5qNSOJgaJpZM4IalSN
.

[dlmacedo]

No.

When trying to run plain NN on the CPU with tensor of type float (no cuda or cunn), I got the following error:

/home/dlm/torch/install/bin/luajit: /home/dlm/torch/install/share/lua/5.1/nn/Container.lua:67: 0ms | Step: 0ms
In 3 module of nn.Sequential:
In 4 module of nn.Sequential:
/home/dlm/torch/install/share/lua/5.1/nn/Dropout.lua:20: bad argument #1 to 'resizeAs' (torch.FloatTensor expected, got torch.ByteTensor)
stack traceback:
[C]: in function 'resizeAs'
/home/dlm/torch/install/share/lua/5.1/nn/Dropout.lua:20: in function </home/dlm/torch/install/share/lua/5.1/nn/Dropout.lua:16>
[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function </home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:41>
[C]: in function 'xpcall'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
train.lua:149: in function 'opfunc'
/home/dlm/torch/install/share/lua/5.1/optim/sgd.lua:44: in function 'sgd'
train.lua:158: in function 'train'
train.lua:234: in main chunk
[C]: in function 'dofile'
.../dlm/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670

WARNING: If you see a stack trace below, it doesn't point to the place where this error occured. Please use only the one above.
stack traceback:
[C]: in function 'error'
/home/dlm/torch/install/share/lua/5.1/nn/Container.lua:67: in function 'rethrowErrors'
/home/dlm/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
train.lua:149: in function 'opfunc'
/home/dlm/torch/install/share/lua/5.1/optim/sgd.lua:44: in function 'sgd'
train.lua:158: in function 'train'
train.lua:234: in main chunk
[C]: in function 'dofile'
.../dlm/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670