Skip to content

Latest commit

 

History

History
165 lines (79 loc) · 3.2 KB

README_usage_CLI.md

File metadata and controls

165 lines (79 loc) · 3.2 KB

Comand Line tools :

- ml_models    :  Running model training
- ml_optim     :  Hyper-parameter search
- ml_benchmark :  Benchmark
- ml_test      :  Testing for developpers.


path are relative the the install folder of mlmodels:
     dataset/timeseries/myfile.csv mean  <Install Folder mlmodels>/dataset/timeseries/myfile.csv



How to use Command Line

ml_models

setup, fit, predict, save, load a model

ml_models --do  
    init            : copy to path  --path "myPath"
    generate_config : generate config file from code source
    model_list      : list all models in the repo
    fit             : wrap fit generic method
    predict         : predict  using a pre-trained model and some data
    test            : Test a model

    
#### Examples

### Copy Notebooks to path
ml_models --do init  --path ztest/

### list all models available in the repo
ml_models --do model_list  


#### generate JSON config file for one model
ml_models  --do generate_config  --model_uri model_tf.1_lstm  --save_folder "ztest/"


#### Fit model and Save
ml_models --do fit     --config_file model_tf/1_lstm.json --config_mode "test"


#### Load model and Save results
ml_models --do predict --config_file model_tf/1_lstm.json --config_mode "test"



####  Internal model
ml_models  --do test  --model_uri model_tf.1_lstm


#### Other examples
ml_models --do fit  --config_file dataset/json/benchmark_timeseries/gluonts_m4.json --config_mode "deepar"

ml_models --do fit  --config_file dataset/json/benchmark_timeseries/gluonts_m5.json --config_mode "deepar"





#### External  Models by Absolute path URI
ml_models --do test  --model_uri "example/custom_model/1_lstm.py"



ml_optim

Hyper-parameter search

ml_optim --do
    test      :  Test the hyperparameter optimization for a specific model
    test_all  :  TODO, Test all
    search    :  search for the best hyperparameters of a specific model


#### For normal optimization search method
    ml_optim --do search  --config_file template/optim_config.json  --config_mode "test"


###### for pruning method
    ml_optim --do search  --config_file template/optim_config_prune.json   --config_mode "test"


###### Using Model default params
    ml_optim --do test   --model_uri model_tf.1_lstm   --ntrials 2



ml_benchmark

## Benchmark model

#### One Single file for all models
ml_benchmark  --do  dataset/json/benchmark.json  --path_json  dataset/json/benchmark_timeseries/test02/model_list.json
     

#### Many json                            
ml_benchmark  --do  dataset/json/benchmark.json  --path_json  dataset/json/benchmark_timeseries/test01/



ml_distributed : Distributed training on Pytorch Horovod

### Work in Progress

#### Distributed Pytorch on CPU (using Horovod and MPI on Linux, 4 processes)  in model_tch/mlp.py
    mlmodels/distri_torch_mpirun.sh   4    model_tch.mlp    mymodel.json

    ml_distributed  --do fit   --n_node 4    --model_uri model_tch.mlp    --model_json mymodel.json



ml_test







Example of CLI in Colab :


https://colab.research.google.com/drive/1u6ZUrBExDY9Jr6HA7kKutVKoP5RQfvRi#scrollTo=4qtLQiaCaDaU