nllLoss lib files and test files #49
Conversation
| x[1,0] = -0.5; x[1,1] = -1.5; x[1,2] = -2.5; | ||
|
|
||
| var target = Tensor.arange(2); | ||
| ref y = target.forceRank(1); |
|
|
||
| var weight = Tensor.ones(3); | ||
|
|
||
| var a = nllLoss(input, target, weight); |
There was a problem hiding this comment.
Should be dynamicTensor.nllLoss(...)?
lib/DynamicTensor.chpl
Outdated
| input: dynamicTensor(2,?eltType), | ||
| target: dynamicTensor(1,eltType), | ||
| weight: dynamicTensor(1, eltType), |
There was a problem hiding this comment.
dynamicTensor does not have a rank argument.
lib/StaticTensor.chpl
Outdated
| proc type staticTensor.nllLoss( | ||
| input: staticTensor(2,?eltType), | ||
| target: staticTensor(1,eltType), | ||
| weight: staticTensor(1, eltType) = ones(), |
There was a problem hiding this comment.
staticTensor.ones(eltType) or make a function overload that doesn't have the weight argument and one that does have the weight argument, but with no default parameter.
I implemented the changes. A bunch of other issues came up too, and now I'm at the point where it's returning this.🌱 loss/nll
|
| @@ -653,6 +653,44 @@ proc dynamicTensor.flatten(): dynamicTensor(eltType) { | |||
| return new dynamicTensor(eltType); | |||
| } | |||
|
|
|||
| proc type dynamicTensor.nllLoss( | |||
| input: dynamicTensor(?eltType), | |||
| target: dynamicTensor(int), | |||
There was a problem hiding this comment.
Should be target: dynamicTensor(eltType)
| for param rankIn in 2..2 { | ||
| if input.checkRank(rankIn) { | ||
| for param rank in 1..1 { | ||
| if target.checkRank(rank) && weight.checkRank(rank) { |
There was a problem hiding this comment.
Should be if target.checkRank(rankIn) && weight.checkRank(rank) {...
| for param rankIn in 2..2 { | ||
| if input.checkRank(rankIn) { | ||
| for param rank in 1..1 { | ||
| if target.checkRank(rank) { |
There was a problem hiding this comment.
Should be if target.checkRank(rankIn) {...
Debugged nllLoss in NDArray, Autograd, StaticTensor, and Dynamic Tensor. Started writing tests in test files. Currently getting this error. Might be something wrong with the test file, but I'm not sure.
🌱 loss/nll
Failed to compile loss/nll/nll.chpl
/nfs/stak/users/sadighia/capstone/ChAI/lib/NDArray.chpl:3: warning: an implicit module named 'NDArray' is being introduced to contain file-scope code
/nfs/stak/users/sadighia/capstone/ChAI/lib/StaticTensor.chpl:61: In method 'this':
/nfs/stak/users/sadighia/capstone/ChAI/lib/StaticTensor.chpl:62: error: unresolved call 'staticTensor(1,real(32)).slice(int(64))'
/nfs/stak/users/sadighia/capstone/ChAI/lib/StaticTensor.chpl:346: note: this candidate did not match: staticTensor.slice(dom: domain(?))
/nfs/stak/users/sadighia/capstone/ChAI/lib/StaticTensor.chpl:62: note: because actual argument #1 with type 'int(64)'
/nfs/stak/users/sadighia/capstone/ChAI/lib/StaticTensor.chpl:346: note: is passed to formal 'dom: domain(?)'
/nfs/stak/users/sadighia/capstone/ChAI/lib/StaticTensor.chpl:62: note: other candidates are:
/nfs/stak/users/sadighia/capstone/ChAI/lib/StaticTensor.chpl:351: note: staticTensor.slice(rngs: range ...rank)
/nfs/stak/users/sadighia/capstone/ChAI/lib/Utilities.chpl:344: note: _tuple.slice(param low: int, param high: int)
note: and 6 other candidates, use --print-all-candidates to see them
/nfs/stak/users/sadighia/capstone/ChAI/test/correspondence/loss/nll/nll.chpl:10: called as (staticTensor(1,real(32))).this(args(0): int(64))
⛔ loss/nll
Failed Chapel compilations tests: ['nll']
Failed Python tests: []
Tests to fix: []