【pytorch】nni微调pytorch mnist

1，文件配置

rose@D370:~/mnist$ tree
.
├── config.yml           # 微调配置文件
├── mnist.py             # 启动文件
└── search_space.json    # 参数配置文件

2，启动与停止

nnictl create --config config.yml  # 启动 -p 8888 端口，可选项
nnictl stop  # 停止

启动后，窗口查看

启动不是直接运行mnist.py这个文件，使用nnictl命令

3，配置文件详解
config.yml，此文件配置nni选项

# 作者
authorName: default
# 项目名称
experimentName: example_mnist_pytorch
trialConcurrency: 1
# 最大持续时间
maxExecDuration: 1h
# 最大尝试数量
maxTrialNum: 10
#choice: local, remote, pai
trainingServicePlatform: local
searchSpacePath: search_space.json
#choice: true, false
useAnnotation: false
tuner:
  #choice: TPE, Random, Anneal, Evolution, BatchTuner, MetisTuner, GPTuner
  #SMAC (SMAC should be installed through nnictl)
  builtinTunerName: TPE
  classArgs:
    #choice: maximize, minimize
    optimize_mode: maximize
trial:
  # 启动文件
  command: python3 mnist.py
  codeDir: .
  # GPU数量
  gpuNum: 0

search_space.json，此文件配置微调的最优参数

{
	"optimizer":{"_type":"choice", "_value":["SGD", "Adadelta", "Adagrad", "Adam", "Adamax"]},
	"learning_rate":{"_type":"choice","_value":[0.0001, 0.001, 0.01, 0.1]},
    "dropout_rate":{"_type":"uniform","_value":[0.5, 0.9]},
    "conv_size":{"_type":"choice","_value":[2,3,5,7]},
    "hidden_size":{"_type":"choice","_value":[124, 512, 1024]},
    "batch_size": {"_type":"choice", "_value": [1, 4, 8, 16, 32]}
}