I have some questions about how to deal with 0 value in 2d array. I am planning to work on regression 1d CNN, but most of the data is made of 0. I would like to replace this 0 to other small values such as 0.000001 simply because, if I put 0, regression process would not take into account the dataset values and find what is optimized model for my data set. I have some instinct but, I need to prove it by referencing. Would you be kind enough as to recommend me some references or any idea of this? Thank you for reading.
Asked
Active
Viewed 46 times
1
Akshat Zala
- 710
- 1
- 8
- 23
손상우
- 111
- 1
- 9
-
If using pandas, you could fill all zero value with the mean and this can cancel out any bias. – leopardxpreload Jun 16 '20 at 04:08
1 Answers
1
The easiest thing to do is to jitter your data by adding 0.0001 * numpy.raodom.randn(*your array shape*) to your data. Whether it will help you in your specific problem, no one can tell, of course.
Igor Rivin
- 4,632
- 2
- 23
- 35
-
i can replace it. but do you have some reference of why we are doing this? – 손상우 Jun 16 '20 at 06:51