parameters() uses torch.type instead of type

[nicholas-leonard]

This allows you to pass custom tensors created with torch.class as parameters.

[pavanky]

There are other places torch.type(foo) == 'table' is being used.

$ grep -e "torch.type(" *.lua | grep table
Concat.lua:      if (torch.type(v) == "table") then
Concat.lua:   local isTable = torch.type(input) == 'table'
Concat.lua:   local wasTable = torch.type(self.gradInput) == 'table'
Concat.lua:         if torch.type(currentGradInput) ~= 'table' then
ConcatTable.lua:      if (torch.type(v) == "table") then
ConcatTable.lua:   local isTable = torch.type(input) == 'table'
ConcatTable.lua:   local wasTable = torch.type(self.gradInput) == 'table'
ConcatTable.lua:         if torch.type(currentGradInput) ~= 'table' then
DontCast.lua:   if torch.type(src) == 'table' then
DontCast.lua:      dst = (torch.type(dst) == 'table') and dst or {}
DontCast.lua:      dst = (torch.type(dst) == type_str) and dst or torch.getmetatable(type_str).new()
GPU.lua:   elseif torch.type(proto) == 'table' then
GPU.lua:   if torch.type(src) == 'table' then
GPU.lua:      dst = torch.type(dst) == 'table' and dst or {}
MixtureTable.lua:   if self.table or torch.type(expertInputs) == 'table' then 
Module.lua:    if (torch.type(self.modules) == 'table') then
Module.lua:      if torch.type(from) == 'table' then
NaN.lua:   if torch.type(tensor) == 'table' then
PrintSize.lua:   if torch.type(input) == 'table' then
PrintSize.lua:   if torch.type(gradOutput) == 'table' then
SelectTable.lua:      if (torch.type(v) == "table") then
SpatialUpSamplingBilinear.lua:   if torch.type(params) == 'table' then
test.lua:   if (torch.type(t1) == "table") then
test.lua:   mytester:assert(torch.type(linear.output) == 'table')
utils.lua:   if torch.type(param) == 'table' then
utils.lua:   if torch.type(t2) == 'table' then
utils.lua:      t1 = (torch.type(t1) == 'table') and t1 or {t1}
utils.lua:   if torch.type(t2) == 'table' then
utils.lua:   if torch.type(t2) == 'table' then
utils.lua:      t1 = (torch.type(t1) == 'table') and t1 or {t1}
utils.lua:   if torch.type(t2) == 'table' then
utils.lua:      t1 = (torch.type(t1) == 'table') and t1 or {t1}
View.lua:   return torch.type(self)..'('..table.concat(self.size:totable(), ', ')..')'

I am not sure if all of them require changing to type, but at least some of them do. Can you see if this can be fixed more comprehensively ?

[nicholas-leonard]

This issue replaces a type with torch.type. You are asking for replacing torch.type with type. Typically, the issue is with people using type instead of torch.type to detect tables.

[pavanky]

Ah nevermind the previous comment then, but here's the new search:

$ grep -e "type(" *.lua | grep table | grep -v torch
Bilinear.lua:   assert(input and type(input) == 'table' and #input == 2,
Container.lua:        if type(from) == 'table' then
Container.lua:         elseif type(self[f]) == 'table' then
DontCast.lua:   if type(src) == 'table' then
FlattenTable.lua:  if type(input) == 'table' then
FlattenTable.lua:  if type(input) == 'table' then
FlattenTable.lua:    if type(input_map) ~= 'table' then
FlattenTable.lua:  if type(input_map) == 'table' then
FlattenTable.lua:  assert(type(input) == 'table', 'input must be a table')
FlattenTable.lua:  assert(type(input) == 'table', 'input must be a table')
FlattenTable.lua:  assert(type(input) == 'table', 'gradOutput must be a table')
hessian.lua:         if type(from) == 'table' then
Identity.lua:         elseif type(self[f]) == 'table' then
IndexLinear.lua:   assert(type(input) == 'table')
IndexLinear.lua:   if type(keys) == 'table' and type(values) == 'table' then
SparseLinear.lua:   if type(input) == 'table' then
SpatialFullConvolution.lua:  if type(inputTensor) == 'table' then
SpatialFullConvolution.lua:    if type(inputTensor) == 'table' then
SpatialFullConvolution.lua:      if type(self.gradInput) == 'table' then
SpatialFullConvolution.lua:    if type(input) == 'table' then
SpatialFullConvolution.lua:  if type(inputTensor) == 'table' then
utils.lua:   if #arg > 0 and type(arg[1]) == 'table' then
utils.lua:         elseif type(self[f]) == 'table' then
VolumetricFullConvolution.lua:  if type(inputTensor) == 'table' then
VolumetricFullConvolution.lua:    if type(inputTensor) == 'table' then
VolumetricFullConvolution.lua:      if type(self.gradInput) == 'table' then
VolumetricFullConvolution.lua:    if type(input) == 'table' then
VolumetricFullConvolution.lua:  if type(inputTensor) == 'table' then

[nicholas-leonard]

@pavanky I changed all of them to torch.type except for one.