I am new to Lua and I am trying to learn it. Recently I came across below line of code which I believe is extracting some value from the table.
local context = context_proto[{{1, batch_size}, {1, source_l*imgH}}]
I have not seen this particular approach to read the table before. I would highly appreciate if anyone can help me understand what exactly above code is doing.
The code you see here doesn't make too much sense in native Lua without further code. It is commonly used in Torch. I found your snippet in a torch related script online. So I guess that's a valid guess.
I'm not very experienced with Torch, but from what I see in the documentation this will give you a sub-tensor of context_proto. row 1-batchSize and col source_l * imgH.
I think it is called slicing and it is covered in the following demo/tutorial: https://github.com/torch/demos/blob/master/tensors/slicing.lua
print 'more complex slicing can be done using the [{}] operator'
print 'this operator lets you specify one list/number per dimension'
print 'for example, t2 is a 2-dimensional tensor, therefore'
print 'we should pass 2 lists/numbers to the [{}] operator:'
print ''
t2_slice1 = t2[{ {},2 }]
t2_slice2 = t2[{ 2,{} }] -- equivalent to t2[2]
t2_slice3 = t2[{ {2},{} }]
t2_slice4 = t2[{ {1,3},{3,4} }]
t2_slice5 = t2[{ {3},{4} }]
t2_slice6 = t2[{ 3,4 }]
...
Please refer to the torch documentation for more details.
From the Lua Documentation:
The type table implements associative arrays, that is, arrays that can be indexed not only with numbers, but with any Lua value except nil and NaN.
That code is using a table as an index into another table. It might be clearer if it were written as follows:
local contextIndex = {{1, batch_size}, {1, source_l*imgH}}
local context = context_proto[contextIndex]
nil without using __index metamethod (as the table value used as index is constructed on the fly and can't match anything).
getmetatable(context_proto).__indexwill be called with{{1, batch_size}, {1, source_l*imgH}}as second parameter, and the result of this call will be assigned tocontext