search

ec-jrc / GHS-S2Net

Other
8 stars 5 forks source link

Bugs in GHS_S2_run_CNN #7

Open dineshbkumar06 opened 1 year ago

dineshbkumar06 commented 1 year ago

Dear,

I am using your code to train new models for my study area. I have noticed that there are some bugs in the code for example in line 267 the code tries to perform ModelChecKpoint for fnmodel. But in the code the fnmodel is assigned later. This is a simple issue I could just assign the fnmodel before this line or comment this line out all together.

Both let me run the training and create a CNN model. What I noticed is that when I use the two CNN models to do prediction I see a big difference in the accuracy of the prediction.

I am wondering do you have an update version of the code that you could share?

Thanks and regards, Dinesh

misterspook commented 1 year ago

Hi,

you can use a constant seed value in your workflow and then compare the results. I guess they will be a bit different due to the stochastic nature of the learning process, SGD.