Spatial Low Rank Nonnegative Tensor Factorization Unmixing
- Download miniconda
- Run a miniconda prompt
- conda create -n tf
- conda activate tf
- conda install tensorflow
- This will install tensorflow 2.3 and all required dependencies at the correct versions including a python environment, intel mkl libraries, numpy, etc.
- conda install matplotlib
- At this point you can run the tensor factorizations for the included datasets: h01-samson, h02-jasper, and h03-urban by just typing: python runall.py. It will run in parallel on a multicore CPU.
In order to take advantage on a GPU you need to install tensorflow-gpu with tensorflow 2.0 or greater. Only Nvidia GPUs are supported and you will need the correct version of the cuda toolkit and driver. If Nvidia libraries or the CUDA runtime is not properly setup tensorflow will default to running on the CPU. If a GPU is detected, tensorflow will load the appropriate library and use it.
- conda create -n tf-gpu
- conda activate tf-gpu
- conda install tensorflow-gpu
- conda install matplotlib
- On either environment run: conda install jupyter
- A shortcut is installed on the Start menu that will launch the Jupyter Server and a browser screen pointing to it at: http://loacalhost:8888
- Browse for demo.ipynb
- Click on Kernel->Restart and Run All
Last modified 1/10/2022