Skip to content

Latest commit

 

History

History
83 lines (66 loc) · 6.92 KB

README.md

File metadata and controls

83 lines (66 loc) · 6.92 KB
Raw

GDMix Workflow

GDMix-workflow is a workflow generation toolkit to orchestrate training jobs of the GDMix, which is a framework to train non-linear fixed effect and random effect models. By providing a GDMix config, GDMix-workflow can generate jobs and run them directly, or generate a YAML file that can be uploaded to Kubeflow Pipeline to run training job distributedly on Kubernetes cluster with Kubeflow Pipeline deployed.

Configuration

GDMix-workflow supports two modes, single_node and distributed. For the single_node mode, user will need to install the gdmix-workflow package and spark, GDMix-workflow will prepare jobs and run them on the node. For the distributed mode, GDMix-workflow generates a YAML file that can be uploaded to Kubeflow Pipeline UI. We'll explain more about distributed mode in the section Run on Kubernetes. Once the gdmix-workflow package is installed (pip install gdmix-workflow), user can call

python -m gdmixworkflow.main

plus following parameters:

  • --config_path: path to a gdmix config. Required.
  • --mode: distributed or single_node. Required, default is single_node.
  • --jar_path: path to the gdmix-data jar for GDMix processing intermediate data.
  • --workflow_name: name for the generated zip file to upload to Kubeflow Pipeline. Required by distributed mode only.
  • --namespace: Kubernetes namespace. Required by distributed mode only.
  • --secret_name: secret name to access storage. Required by distributed mode only.
  • --image: image used to launch gdmix jobs on Kubernetes. Required by distributed mode only.
  • --service_account: service account for the spark-on-k8s-operator to launch spark job. Required by distributed mode only.

Run GDMix workflow on Kubernetes for distributed training

GDMix's distributed training is based on Kubernetes, and leverages Kubernetes job scheduling services Kubeflow and spark-on-k8s-operator to run TensorFlow and Spark job distributedly on Kubernetes, and uses Kubeflow Pipeline to orchestrate jobs. Besides that, a centralized storage is needed for storing training data and models. User can use Kubernetes-HDFS or NFS as the centralized storage.

Create a Kubernetes cluster, deploy required services

To run GDMix in the distributed mode, user needs to create a Kubernetes cluster, and deploy following services:

Generate task YAML file and upload to Kubeflow Pipeline UI

When the Kubernetes cluster and services are ready, with the provided GDMix config, GDMix-workflow can generate task YAML file that has job specification for the distributed TensorFlow and Spark jobs. User needs to upload it to Kubeflow Pipeline and start training.

Run the MovieLens example

In this section we'll introduce how to train fixed effect and random effect models using GDMix for MovieLens data. Please download and preprocess moveLens data to meet GDMix's need using the provided script download_process_movieLens_data.py:

wget https://raw.githubusercontent.com/linkedin/gdmix/master/scripts/download_process_movieLens_data.py
pip install pandas
python download_process_movieLens_data.py

For distributed training, the processed movieLens data need to be copied to the centralized storage.

We'll also need a GDMix config, a reference of training logistic regression models for the fixed effect global and the random effects per-user and per-movie with distributed training can be found at lr-movieLens.yaml.

Run on single node

Please see the section Try out the movieLens example in the root README.md for details of how to run the movieLens example on single node.

Run on Kubernetes for distributed training

To run on Kubernetes, as mentioned earlier, user will need to copy the processed movieLens data to the centralized storage, modify the input path fields such as training_data_dir, validation_data_dir, feature_file and metadata_file of the GDMix config for distributed training lr-movieLens.yaml.

If using the provided image linkedin/gdmix, user can mount the processed movieLens data from the centralized storage to path /workspace/notebook/movieLens for each worker, then no change is needed for the distributed training GDMix config lr-movieLens.yaml.

Generate YAML file

User will need to install the GDMix-worklfow package to generate the YAML file:

pip install gdmix-workflow

Download the example GDMix config for distributed training and generate the YAML file with following command:

wget https://raw.githubusercontent.com/linkedin/gdmix/master/gdmix-workflow/examples/movielens-100k/lr-movieLens.yaml

python -m gdmixworkflow.main --config_path lr-movieLens.yaml --mode=distributed --workflow_name=movieLens --namespace=default --secret_name default --image linkedin:gdmix --service_account default

Parameters of namespace, secret_name and service_account relate to the Kubernetes cluster setting and job scheduling operator deployments. A zip file named movieLens.zip is expected to be produced.

Upload to Kubeflow Pipeline

If the Kubeflow Pipeline is successfully deployed, use can forward the Pipeline UI to local browser, The command below forwards the Pipeline UI to the local port 9980:

kubectl -n default port-forward svc/ml-pipeline-ui 9980:80

Type localhost:9980 in the local browser to view the Kubeflow Pipeline UI, upload the produced YAML file movieLens.zip(click button Upload pipeline), and then click button Create run to start the training. A snapshot of the movieLens workflow is shown below.


GDMix Distributed Training on Kubeflow Pipeline