# This file is generated by `ray project create`.

name: ray-example-resnet

description: "Using ray to train resnet on multiple gpus"
tags: ["ray-example", "machine-learning", "tensorflow", "resnet"]
documentation: https://ray.readthedocs.io/en/latest/auto_examples/plot_resnet.html

cluster: .rayproject/cluster.yaml

environment:
  requirements: .rayproject/requirements.txt

commands:
  - name: train
    command: |
      if [ "{{dataset}}" == "cifar10" ]; then
        # Get the CIFAR-10 dataset.
        curl -o cifar-10-binary.tar.gz https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz;
        tar -xvf cifar-10-binary.tar.gz;
      else
        # Get the CIFAR-100 dataset.
        curl -o cifar-100-binary.tar.gz https://www.cs.toronto.edu/~kriz/cifar-100-binary.tar.gz;
        tar -xvf cifar-100-binary.tar.gz;
      fi
      python resnet_main.py --dataset {{dataset}} --train_data_path {{train_data_path}} --eval_data_path {{eval_data_path}} --eval_dir {{eval_data_path}} --eval_batch_count {{eval_batch_count}} --num_gpus {{num_gpus}}
    params:
      - name: dataset
        help: "The dataset to train on."
        default: "cifar10"
        choices: ["cifar10", "cifar100"]
      - name: train-data-path
        help: "Data path for the training data."
        default: "'cifar-10-batches-bin/data_batch*'"
        type: str
      - name: eval-data-path
        help: "Data path for the testing data."
        default: "cifar-10-batches-bin/test_batch.bin"
        type: str
      - name: eval-dir
        help: "Data path for the tensorboard logs."
        default: "/tmp/resnet-model/eval"
        type: str
      - name: eval-batch-count
        help: "Number of batches to evaluate over."
        default: 50
        type: int
      - name: num-gpus
        help: "Number of GPUs to use for training."
        default: 0
        type: int
    config:
      tmux: true

output_files: [
  # Save the logs from the latest run in snapshots.
  "/tmp/ray/session_latest/logs"
]