FAQ Machine Learning
Q: How do I use ML?
A: Please refer to the documentation here.
Q: I want to use the ‘old’ NeuralNetworks plugin. Where can I find it?
A: In OpendTect 6.6, the ‘old’ NeuralNetworks plugin is now nestled inside the Machine Learning Control Center. It functions in exactly the same way as did the standalone NN plugin in OpendTect 6.4 and previous versions.
Q: Is there an overview page of all the OpendTect ML knowledge?
A: Yes, please visit the following page on which one can find links to the OpendTect 6.6 installer, links to free datasets, video links, links to documentation (user, development and workflows, a link to this FAQ and links to code examples and data:
Q: Is it possible to develop your own Machine Learning models?
A: Yes, this is possible. You can read documentation online, view webinar videos and download example code from GitHub:
- Develop your own models documentation
- Machine Learning Webinars: Part 5: Creating and Adding New Models
- OpendTect Webinar: Develop your own Machine Learning tools and workflows with OpendTect
- OpendTect Webinar: How to prepare well logs to get optimal Machine Learning results
- OpendTect-ML-Dev GitHub repository
Q: To develop in OpendTect - ML, do I need a license for OpendTect Pro and the Machine Learning plugin?
A: No, you do need to install OpendTect Pro and Machine Learning but you can develop new Machine Learning models and workflows that can be tested on a number of free datasets that can be downloaded from our TerraNubis portal. These special datasets do not check for licenses.
Q: Can I test my models free-of-charge on my own datasets?
A: Yes, this is possible if you are willing to release the datasets under the Creative Commons License via TerraNubis. If you cannot share the data, you need a license for OpendTect Pro and Machine Learning. Commercial users can get licenses from the OpendTect Pro store. Universities can get free licenses under our Academic License Agreement (apply here).
Q: What are my options for sharing trained models?
A: Our goal is to build a world-class library with trained models that can be imported into OpendTect - Machine Learning so users can apply these to solve similar problems on their own datasets. You have complete freedom to decide if you want to share your trained model free-of-charge and in the public-domain, licensed on your own defined commercial terms, or to keep your model proprietary. If the model will have a license, please be sure to include that in a license text file included with your submissions. Models without licenses will be assumed to be in the public-domain.
Q: Why does the GPU not use all its resources during certain Machine Learning processes?
A: Within the Machine Learning plugin, it is the Python application that is running either the training or the prediction, not OpendTect itself. Within Python, the performance and behavior of each process (training/prediction) depends entirely on the python module being used: It will be very different between Sklearn (CPU only, very small memory footprint), and Tensorflow (GPU or CPU, large memory utilization).
We keep monitoring any available update for these Python packages and will implement these newer, improved versions immediately as they become available.