I figured out what the problem was. T.arange creates int64's by default and when operated on by a float, they both get upcasted to float64 and not floatX specified in config. Maybe this will be useful for some one else in the future.
On Saturday, September 10, 2016 at 7:56:20 PM UTC+5:30, Pranav Shyam wrote: > > Hi, > > I have been trying code a draw like attention model for a project. The > code is not compiling with floatX set to float32 due a dtype conflict. It > says that some operation inside scan is causing the result to be of float64 > type but my outputs_info are initialized as float32. I checked everything > multiple times. All numpy arrays are in float32, and I couldn't find any > area where upcasting to float64 could occur. Can someone please help me > with this and tell me what exactly is causing the error? > > Note: I have reduced code from multiple files to this one file which is > about 135 lines of code. > > Also, this works okay when floatX flag is set to float64. > > Thanks and Regards, > Pranav > -- --- 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.
