numpy 1.23.5
tensorflow 2.12.0
You need to have tensorflow installed. Refer to the official documentation and install tensorflow according to your system. If your system don't have a CUDA supported GPU, you might need to install tensorflow-cpu instead.
The trained model on both BLO and LOW is in trained_models/all40. A function that use it is in predict_cloudy.py, and you need to put a single image in a seperate folder, for example:
main_directory/image_to_clasify.png
Then give the function the path to the folder.
As for why you need a seperate folder, it is because of a weird bug here and here
Try
python predict_cloudy.py
To check if the function can run or not
modify fetch-images.py to set the site and the dates to pull from and the output folder
python3 fetch-images.py
to generate cmd-list.sh
Then you can bash cmd-list.sh
to start dowloading
dir /data/blo/2022 >> text.txt
Put text.txt onto remote2 and run Cloud sensor data for ML training notebook. Besure to set the site in get_sub_temp_log. It will read from text.txt and output is_cloudy.csv
Download is_cloudy.csv to local machine and run sort_date_into_folders.ipynb
Be sure to change folder locations and blacklisat dates
Modify train.py to specify data folders and run python3 train.py
or python3 trainmixed.py
(that train data on multiple data)
It will output a folder of checkpoints for each epoch so you can continue training from an epoch without training from beginning over again