After running some tests, it seems like:

- Use `add` where the input can be an arbitrarily long list of variables, 
and we want to add them like x+y+z+...
- Use `sum` when we have some matrix/vector and we want to sum its 
components.

On Monday, April 24, 2017 at 9:14:07 AM UTC-7, Daniel Seita wrote:
>
> Hmmm ... upon reading the same docs page linked above, I see the following:
>
> __{abs,neg,lt,le,gt,ge,invert,and,or,add,sub,mul,div,truediv,floordiv}__
>
> Those elemwise operation are supported via Python syntax.
>
> That must be where `T.add` comes from. I'll still need to investigate on 
> the difference between `T.add` and `T.sum` but maybe this will help...
>
> On Monday, April 24, 2017 at 9:09:38 AM UTC-7, Daniel Seita wrote:
>>
>> Hi all,
>>
>> I am reading through some theano code written by someone else (the 
>> actual file is here 
>> <https://github.com/joschu/modular_rl/blob/d1aeb866f4293d4458de1bc02a42e380289e9c43/modular_rl/core.py>
>>  
>> if you're curious but it shouldn't matter for this question).
>>
>> I've noticed that the code uses two functions from theano: `add` and 
>> `sum`.
>>
>> I found the theano.tensor.sum docs linked here 
>> <http://deeplearning.net/software/theano/library/tensor/basic.html?highlight=theano%20tensor%20sum#theano.tensor.sum>.
>>  
>> However, I surprisingly can't find any documentation for the `add` method 
>> (or even any existing questions here or on stack overflow). I know it seems 
>> obvious that the add method just sums up its arguments, but I just want to 
>> be extra sure. I'm also curious about if there's an alias, i.e. if 
>> `T.add(x,y)` is the same as `x+y` where `x` and `y` are theano variables. 
>> Finally, I'd also be curious about the differences between `add` and `sum` 
>> in theano.
>>
>> Thanks.
>>
>

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to