ClassifyPixels crash

Hi dear CellProfilers,

I am trying to run ClassifyPixels but I do run into either
one of the two error messages:

Traceback (most recent call last):

File
“cellprofiler\gui\pipelinecontroller.pyc”, line 2777, in do_step

File
“cellprofiler\modules\classifypixels.pyc”, line 244, in run

File
“cellprofiler\modules\classifypixels.pyc”, line 272, in
get_classifiers

File
“cellprofiler\modules\classifypixels.pyc”, line 303, in
parse_classifier_file

File
“cellprofiler\modules\classifypixels.pyc”, line 319, in
parse_classifier_hdf5

File
“h5py\highlevel.pyc”, line 793, in init

File
“h5py\highlevel.pyc”, line 827, in _generate_fid

File
“h5f.pyx”, line 68, in h5py.h5f.open (h5py\h5f.c:1264)

FileError: File close degree doesn’t match (File
accessability: Unable to initialize object)

Traceback (most recent call last):

File
“cellprofiler\gui\pipelinecontroller.pyc”, line 2777, in do_step

File
“cellprofiler\modules\classifypixels.pyc”, line 244, in run

File “cellprofiler\modules\classifypixels.pyc”,
line 272, in get_classifiers

File
“cellprofiler\modules\classifypixels.pyc”, line 303, in
parse_classifier_file

File
“cellprofiler\modules\classifypixels.pyc”, line 335, in
parse_classifier_hdf5

AttributeError: type object ‘ClassifierRandomForest’ has no
attribute ‘loadRFfromFile’

Maybe you could give me a suggestion to work around this
problem?

the classifier file was generated using the random forest
classifier with variable importance. For now I am running the pipeline in
“test mode” stopping it at the “IdentifyPrimaryObjects”
step.

The required files are uploaded below:
LIP_ilastik0.5_cells_only_clipped.h5.zip (197.0 KB)
CellProfilerTestMicropattern v5.1 LIP.cpproj (643.0 KB)

test images will follow! Thanks!

I have attached a drastically compacted pipeline and a demo image for the pipeline:
bugdemo_NBE.cpproj (401.4 KB)
1_EGF_EGFR_cam2_AutoJoin_cropped.zip (709.2 KB)

Found the problem/bug: The random forest classifier with variable importance caused the crash. If I train with the standard random forest algorithm, ClassifyPixels works just fine.
As I understand, the random forest classifier with variable importance is superior with objects of variable brightness (e.g. varying expression levels of GFP-tagged protein)? Or to apply the same classifier on images on different setup with different intensities/signals? Or is this a misunderstanding?

have a nice weekend!
Benedikt

edit: I do understand now that the variable importance is there for the analysis of the importance of features. Hope I got that right :wink: