Hi @VHolubec ,
If you want to use a better GPU for training a deep learning model, we provide ArcGIS Notebook Server with a ready to use deep learning environment.
Credentials: https://geosaurus.maps.arcgis.com/home , arcgis_python, P@ssword123
You need to create a notebook with 'Advanced with GPU support' option as shown in the screenshot.
Now coming to how will you get your exported data on this portal, please refer to https://developers.arcgis.com/python/sample-notebooks/vehicle-detection-and-tracking/ or any other deep learning notebook available. You can do the following steps:
1. zip your training data and upload it as image collection item. For that, click on Contents> Add item>from your computer. Then upload zip file and change the type to image collection.
2. Follow the steps from the notebook link mentioned above and unzip your data and complete the rest of the workflow to train a model. When you get a good model, call save('name', publish=True). This will publish the dlpk as an item and you can download it locally to your disk.
Let me know if there's anything we could help with.
Regards,
Priyanka
The main reason why I would use the notebook for the AGP workflow is due to the vanilla training model tool simply lacking compatibility with the demands of the data. I am at this moment training a model using jupyter notebook in browser while importing arcgis learn module. So I am getting it to work, with this work around, though of course I want to develop a repeatable workflow for just using the internal AGP notebook.
So I am about 100% sure that the reason why the AGP notebook was not working correctly (ie., displaying batches), was due to some packages requiring updates- 41 in total, apparently (see attached image). However, as I mentioned before, I cannot seem to update any of them inside of the AGP options interface at all. Like I click "update" and it does nothing. I've tried doing that and restarting it, and nothing changes. This is with a cloned environment where it should work. I've also tried changing permissions on the cloned environment file itself, but it seems to be at highest level already. I also tried updating it within the python command shell, but while it says it will do all of the package updates, it also requires a few package downgrades- one of which was the jupyter server or whatever, which crashed notebooks inside of AGP.
So if I understand you correctly, are you saying that I could just update a specific package in the cloned environment using the python command shell? That may work, just seems really inefficient. The part that I don't understand is why is required the downgrade for some packages, which seemed odd.
EDIT: just to clarify, arcgis 1.8.5 is installed in my environments.
Hello @Anonymous User ,
so just to understand your status.
1) You have created a cloned Python environment in ArcGIS PRO
2) You have set it up as default
3) You have installed there some other python packages
Is that correct?
To be honest I don´t use the PRO for managing the Python. So I can give you an advice to manage it with Python Command prompt.
Open Python Command prompt
There verify the environment which is used - in my case it´s cloned environment called DL
- to install a package run: conda install <package name>
- to install a package with a specific version run: conda install <package name>=<version>
- to update a specific package run: conda update <package name>
- to update all run: conda update --all
The last command will update all packages in the current environment to the latest version, but it may use an older version of some packages in order to satisfy dependency constraints.
Hope this helps.
Vladimir