Port trial examples' config file to v2 #3721
Conversation
|
Add example for config_hybrid.yml and config_windows_v2.yml? |
I added a comment line in every "basic" example to inform windows users. I think a separate Windows example is bad because it makes me think NNI behaves differently on Windows and Linux. Let's discuss it in the meeting. |
| @@ -0,0 +1,15 @@ | |||
| searchSpaceFile: search_net.json | |||
| trialCodeDirectory: EfficientNet-PyTorch | |||
| trialCommand: python main.py /data/imagenet -j 12 -a efficientnet --batch-size 48 --lr 0.048 --wd 1e-5 --epochs 5 --request-from-nni | |||
| @@ -1,23 +1,14 @@ | |||
| authorName: default | |||
| experimentName: example_pytorch_cifar10 | |||
| searchSpaceFile: search_space.json | |||
There was a problem hiding this comment.
suggest at least for this simple example, we add "experimentName", because if users really use nni to run experiments, they want to give each experiment an easy-to-remember name (not experiment ID). If this field is not added in the example, users have to check config references, which is not friendly
| experimentName: example_mnist_pytorch | ||
| # This is the minimal config file for an NNI experiment. | ||
| # Use "nnictl create --config config.yml" to launch this experiment. | ||
| # Afterwards, you can check "config_detailed.yml" for more explaination. |
There was a problem hiding this comment.
explaination -> explanation
| @@ -0,0 +1,42 @@ | |||
| # This example shows more configurable fields comparing to the minimal "config.yml" | |||
| # You can use "nnictl create --config config_detailed.yml" to launch this experiment. | |||
| # If you see an error message saying "port 8080 is used", use "nnictl stop --all" to stop previous experiment. | |||
There was a problem hiding this comment.
-> use "nnictl stop --port 8080" to stop that experiment, or use "nnictl stop --all" to stop all the previous experiments.
There was a problem hiding this comment.
I prefer not to provide that much details in example. At this point the user does not need to know how to manage multiple experiments.
I'm afraid that "--port 8080" might threaten newbie users.
|
|
||
| trialCommand: python3 mnist.py # The command to launch a trial. NOTE: change "python3" to "python" if you are using Windows. | ||
| trialCodeDirectory: . # The path of trial code. By default it's ".", which means the same directory of this config file. | ||
| trialGpuNumber: 1 # How many GPUs should each trial use. CUDA is required when it's greator than zero. |
| momentum: | ||
| _type: uniform | ||
| _value: [0, 1] | ||
|
|
There was a problem hiding this comment.
yeah, "experimentName" should be put here, we can tell users that they can omit it if they don't want to write it.
|
looks great! |
Notes:
Fixed bugs: