problem of train cnn model

sassoftware / python-dlpy

The SAS Deep Learning Python (DLPy) package provides the high-level Python APIs to deep learning methods in SAS Visual Data Mining and Machine Learning. It allows users to build deep learning models using friendly Keras-like APIs.

Apache License 2.0

224 stars 131 forks source link

problem of train cnn model #240

Open banbsyip opened 5 years ago

banbsyip commented 5 years ago

I set up a cnn model to classify mnist data.the model as follow:

i train the model firstly successful,the result as follow: when i changed the characters ,if the max_epochs greater than step_size, the train go wrong.

ethem-kinginthenorth commented 5 years ago

learning rate schedulers are created for easy support to learning rate modification. we will have checks to prevent a smaller step_size than the max_epochs when lr_schedulers are used.

banbsyip commented 5 years ago

learning rate schedulers are created for easy support to learning rate modification. we will have checks to prevent a smaller step_size than the max_epochs when lr_schedulers are used.

when I trained the tinyyolo, the same problem occurred.the code from https://go.documentation.sas.com/?cdcId=pgmsascdc&cdcVersion=9.4_3.4&docsetId=casdlpg&docsetTarget=p1np8zbnoyd0brn1dhehthuuxj4q.htm&locale=zh-CN#p1dgtuox2j70uen1sg8qpg203n9z it is about yolo's object detection

ethem-kinginthenorth commented 5 years ago

thanks for pointing out. I will be looking into this today.

ethem-kinginthenorth commented 5 years ago

@banbsyip I have been trying to reproduce the issue but I cannot. I feel like there is another issue is happening here.

ethem-kinginthenorth commented 5 years ago

@banbsyip seems like there might be an issue with the connection or such. would you please set "swat.options.cas.debug.responses = True" and then re-run your code. I would like to see what the server sends back

banbsyip commented 5 years ago

eems like there might be an issue with the connection or such. would you please set "swat.options.cas.debug.responses = True" and then re-run your code. I would like to see what the server sends back

ok,i will run the code as your suggestion. I have a question about the label,i use the voc2007 data,the label of "person" have some parts such as "head","left","hand"and so on. I label all the parts as "person", is it a correct way to label?

sasxianhu commented 5 years ago

Depends on what your task is. If you want to detect the parts, then label them for sure. If you think any human parts should use person as label, then label them as person.

banbsyip commented 5 years ago

Depends on what your task is. If you want to detect the parts, then label them for sure. If you think any human parts should use person as label, then label them as person.

thanks for your help.

banbsyip commented 5 years ago

@banbsyip seems like there might be an issue with the connection or such. would you please set "swat.options.cas.debug.responses = True" and then re-run your code. I would like to see what the server sends back the result as follow:

ethem-kinginthenorth commented 5 years ago

@banbsyip thanks for sharing the log message. I am communicating with swat folks to see if they can help resolve this issue.

Guma02 commented 1 year ago

Hello, I know this is an old forum, but I am having the same error message. Does anyone know the solution? @banbsyip @ethem-kinginthenorth Thanks in advance.