This project uses HyperOpt's Bayesian optimization and Ray Tune to perform hyperparameter tuning for a simple image classifier. I explain the code and concepts used in this project in this blogpost.
Install the dependencies by running the command:
$ pip install -r requirements.txt
You can download the dataset using the data.py script. You can also specifiy the number of images to kept in train and validation folder using the --train-count and --val-count flags:
$ python data.py --train-count 500 --val-count 100
This script downloads and extracts the data in the current working directory in ./dataset folder.
Verify that hyperopt and ray is able to execute by running the following test script:
$ python test_hyperopt.py --smoke-test
Once the smoke test is successfull you can proceed to hyperparameter tuning using train.py
Before starting hyperparameter tuning process please update the num_cpus and num_gpus variables in train.py according to your machine configuration. The default values are num_cpus=2 and num_gpus=0. You can then start the hyperparameter tuning process by running the following command:
$ python train.py --train-dir dataset/train --val-dir dataset/val --optimize True --samples 100
If you do not want to run hyperparameter tuning but just want to train the image classifier using the default configuration (specified in default_config.json) then you can use the following command:
$ python train.py --train-dir dataset/train --val-dir dataset/val
Once the training is completed, the artifacts are stored in the ./snapshots directory. You can run the inference by pass the ./test_images folder to the following command:
$ python inference.py --test-dir ./test_images --snapshot-dir ./snapshots
The output of the inference would be a Pandas dataframe printed in the stdout