United-Data-Pioneering

Project #4 - Impervious Surfaces

Using anaconda with python3.6 and tensorflow-gpu 1.5

Setup was a bit of a pain;

conda create --name {envname} python=3.6
conda install ipykernel --name {envname}
python -m ipykernel install --prefix=C:/anaconda/envs/{envname} --name {envname}
activate envname

python -m pip install tensorflow-gpu==1.5

conda install -c anaconda cudatoolkit=9.0

conda install cuDNN=7.6.0

Then with many incompatible dependency versions, I ended up with the package versions exported from pip freeze found in requirements.txt, which can be installed by

python -m pip install -r requirements.txt

I can get to the training step, last I left it it was stuck on output "Epoch 1/12"

checkpoint.h5 too large to upload to github