Error with arcgis.learn.prepare_data on arcgis==2.1.0.2

[Aziz_AlNajjar]

Hello there, 

I've been using the ArcGIS package on my computer using the conda package installations with no problems. However, I needed better computing abilities to train the model faster. So I tried to install the ArcGIS on the graham cluster from Compute Canada which only allows pip installations, and it is prohibited to use conda. 

I used the pip install arcgis==2.1.0.2  to install the library as specified in the installation instructions and there were no problems. 

However, when I try to use the prepare_data function on my exported point cloud data I get the following error: 

" /home/aziz99/ENV/lib/python3.8/site-packages/arcgis/learn/models/_arcgis_model.py:459: UserWarning: Cuda is not available
warnings.warn("Cuda is not available")
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/aziz99/ENV/lib/python3.8/site-packages/arcgis/learn/_data.py", line 2687, in prepare_data
data = pointcloud_prepare_data(
File "/home/aziz99/ENV/lib/python3.8/site-packages/arcgis/learn/_utils/pointcloud_data.py", line 1845, in pointcloud_prepare_data
src=PointCloudItemList.from_folder(
AttributeError: type object 'PointCloudItemList' has no attribute 'from_folder'"

I am pretty sure that there is nothing wrong with the data as it works normally when I use the prepare_data on the conda environment. 

I really appreciate your help, 

Thank you in advance 

[Kepa]

Hello @Aziz_AlNajjar,

I'm affraid a simple pip installation is not enough to run deep learning tools out of the box. There are a lot of dependencies involved. There is even a specific metapackage on conda.

As a workaround you could zip your working conda environment and copy/paste in the target machine (you won't be able to use arcpy if ArcGIS PRO isn't present or if there is a version mismatch, but it seems that you are only using arcgis.learn). However not sure if this will work as expected.

Also notice that you are running tools without CUDA which will increase processing time (even using CPU parallel processing).

Maybe the best option is to talk to TI and try to install a proper conda environment with the whole package set.

Regards,