clinicadl train --visualization Save results in visualization folder about clinicadl HOT 9 CLOSED

Hello，
How to set up dropout during training？

from clinicadl.

Hi,

Where can I view this result? And how to draw ROC curve and calculate AUC?

The visualization folder stores images for one subject in NIfTI format after multiples epochs. Visualizers for NIfTI format can be easily grabbed on the web. To draw the ROC and calculate AUC we suggest you to activate the TensorBoardX module while running the training and follow in a web interface the evolution of the curves.

from clinicadl.

Hello，
How to set up dropout during training？

Hello @921974496,

@14thibea created PR #11 to address your question (currently, this is not possible). Once reviewed, this will be available for AD-DL.

Best,
Alex

from clinicadl.

Hi,
PR #11 was merged to master. Now, it's possible to to set dropout parameter during the training for the concerned layers.

from clinicadl.

Hi,
PR #11 was merged to master. Now, it's possible to to set dropout parameter during the training for the concerned layers.

Thanks

from clinicadl.

When I run clinicidl classify I encounter some problems：
File "/home/geng/AD-DL/clinicadl/clinicadl/tools/deep_learning/iotools.py", line 164, in commandline_to_json
if commandline_arg_dic['split'] is None:
KeyError: 'split'

Is my network_dir parameter wrong? My network_dir parameter is the best_model_dir folder in train output_dir

from clinicadl.

The command I used is clinicadl classify patch / home / geng / ADNI2 / test / caps / home / geng / ADNI2 / test / tsv / home / geng / ADNI2 / test / output / home / geng / ADNI2 / resault / best_model_dir

from clinicadl.

Hi,
PR #11 was merged to master. Now, it's possible to to set dropout parameter during the training for the concerned layers.

The “class Parameters” should also be modified in the “clinicadl / clinicadl / tools / deep_learning / iotools.py file”

from clinicadl.

When I run clinicidl classify I encounter some problems：

Hi,
the classify task is currently WIP (The brand new issue #15 clarifies this).
However, you can try to use one of the scripts available to evaluate your trained model (see clinicadl/clinicadl/subject_level/evaluation.py for example).

from clinicadl.