Hello, In theano 0.8.2, T.nnet.sigmoid implemented a grad method: <https://lh3.googleusercontent.com/-HbNMPmAXf6w/WIuT-0QDh0I/AAAAAAAAFK4/LON8RRdnG0QaS_xoyKI6bCj3zJNgnCzFgCLcB/s1600/Screen%2BShot%2B2017-01-27%2Bat%2B10.21.55%2BAM.png> However, in the bleeding edge version of theano, this method doesn't exist: <https://lh3.googleusercontent.com/-OydJN6Fq_ho/WIuUdA-zaPI/AAAAAAAAFLA/YpvhqdTtFfUbM_mJiGtPBSlXW3LD1CIPwCLcB/s1600/Screen%2BShot%2B2017-01-27%2Bat%2B10.41.31%2BAM.png> And yet, if I use the sigmoid function to compute something and call T.grad on the result, it works: <https://lh3.googleusercontent.com/-CI3QuTfWltE/WIuW_QbTAeI/AAAAAAAAFLY/0lSzHFPibG4GKSU6uxvPdcRDBUShum7vgCLcB/s1600/Screen%2BShot%2B2017-01-27%2Bat%2B10.52.13%2BAM.png> I looked at a number of other ops and it appears they still retain their grad methods. I'm wondering why the grad method was deprecated for Elemwise, and if there's an easy way to access the sigmoid gradient operation in the current version of theano? -- --- 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.
