The appearance of full nan

[anny0316]

Hello Ryan-Rhys, Thanks for you code. I am interested with FlowMo. When I run Gaussian_Process_String_Kernel.ipynb, nan is appeared. I found that the problem is in this sentence: k_results = tf.divide(k_results, tf.sqrt(norm)). Because tf.sqrt(norm)==0. I want to know why there is such a result. Thanks a lot.

Regards anny

[Ryan-Rhys]

Hi Anny,

Thanks for raising this and apologies for the delay in responding! I think tagging either @Ryan-Rhys or @henrymoss should give us a notification in the future!

Handing this one over to @henrymoss since he coded up the string kernel!

Best wishes, Ryan

[Ryan-Rhys]

@henrymoss

[JinhaoGu]

Hello Ryan-Rhys, Thanks for you code. I am interested with FlowMo. When I run Gaussian_Process_String_Kernel.ipynb, nan is appeared. I found that the problem is in this sentence: k_results = tf.divide(k_results, tf.sqrt(norm)). Because tf.sqrt(norm)==0. I want to know why there is such a result. Thanks a lot.

Regards anny

Hi Rayn and Anny,

I had the same problem with the demo. This problem came from the function tf.strings.split when split SMILES strings. Just change tf.strings.split to tf.strings.bytes_split in GP/kernels.py line 118 and line 129. The problem should be solved. @Ryan-Rhys @henrymoss

Best, Jinhao

[Ryan-Rhys]

Hi Jinhao/Anny,

Many thanks for that fix!

Best wishes, Ryan