ml2p 0.1.5

Overview

ML2P helps manage a machine learning project. You’ll define your project by writing a small YAML file named ml2p.yml:

project: "zendesk-complaint-flagger"
s3folder: "s3://zendesk-complaint-flagger-sagemaker-test/"
models:
  zcf: "zendesk_complaint_flagger.ml2p.ZCFModel"
defaults:
  image: "670223297817.dkr.ecr.eu-west-1.amazonaws.com/zendesk-complaint-flagger-sagemaker:latest"
  role: "arn:aws:iam::319756771752:role/service-role/AmazonSageMaker-ExecutionRole-20190410T162807"
train:
  instance_type: "ml.m5.xlarge"
deploy:
  instance_type: "ml.t2.medium"

This specifies:

• the name of your project,

• the docker image that your project will use for training and prediction,

• the S3 bucket that will hold the models and data sets for your project,

• the AWS role your project will run under,

• and the Python class(es) that will be used to train and serve your models.

The name of your project functions as a prefix to the names of SageMaker training jobs, models and endpoints that ML2P creates (since these names are global within a SageMaker account).

ML2P also tags all of the AWS objects it creates with your project name.

Setting up your project

To set up your project you’ll need to create:

• the docker image,

• the S3 bucket,

• and the AWS role

yourself. ML2P does not manage these for you.

When you run ml2p init (see below), ML2P will create the following folder structure in your S3 bucket:

s3://my-project-bucket/
  models/
    ... ml2p with place the outputs of your training jobs here ...
  datasets/
    ... you should place your data sets here, the name of the
        subfolder is the name of the data set ...

Assumptions made by ML2P

ML2P is minimal, opinionated and probably wrong in some cases. It currently assumes:

• You want your datasets and models stored in the folder structure described above.

• You’ve uploaded your training data to folders under datasets/.

Training and deploying a model

First set the AWS profile you’d like to use:

$ export AWS_PROFILE="prodigy-development"

If you haven’t initialized your project before, run:

$ ml2p init

which will create the S3 model and dataset folder for you.

Next start a training job to train your model:

$ ml2p training-job create zcf-train-6 zcf-20190412 --model-type zcf

The first argument is the name of the training job, the second is name of the data set (i.e. the folder under /datasets/ in your project’s S3 bucket). You will need to have uploaded some training data. The –model-type argument is optional – the model type to use may also be specified directly in the docker image.

Wait for your training job to finish. To check up on it you can run:

$ ml2p training-job wait zcf-train-6  # wait for job to finish
$ ml2p training-job describe zcf-train-6  # inspect job

Once your training job is done, create a model from the output of the training job:

$ ml2p model create zcf-model-6 zcf-train-6 --model-type zcf

The first argument is the name of the model to create, the second is the training job the model should be created from. The –model-type argument is optional – the model type to use may also be specified directly in the docker image.

The model is just an object in S3 – it doesn’t run any instances – so it will be created immediately.

Now its time to deploy your model by creating an endpoint for it:

$ ml2p endpoint create zcf-endpoint-6 zcf-model-6

The first argument is the name of the endpoint to create, the second is the name of the model to create the endpoint form.

Setting up the endpoint takes awhile. To check up on it you can run:

$ ml2p endpoint wait zcf-endpoint-6  # wait for endpoint to be ready
$ ml2p endpoint describe zcf-endpoint-6  # inspect endpoint

Once the endpoint is ready, your model is deployed!

You can make a test prediction using:

$ ml2p endpoint invoke zcf-endpoint-6 '{"your": "data"}'

And you’re done!

Working with models locally

At times it may be convenient to work with ML2P models on a local machine, rather than within SageMaker. ML2P supports both training models locally and loading models trained in SageMaker for local analysis.

In either case, first create a local environment:

# set up a connection to AWS, specifying an appropriate AWS profile name:
import boto3
session = boto3.session.Session(profile_name="aws-profile")

# create a local environment, the arguments are the local folder to store the
# environment in, the path the ml2p.yml config file, and an optional boto3
# session to use for retrieving files from S3.
from ml2p.core import LocalEnv
env = LocalEnv("./local", "./sagemaker/ml2p.yml", session)

# import your ml2p model class:
from my_package import MyModel

Then to train a model locally:

env.download_dataset("dataset-name")
env.clean_model_folder()
trainer = MyModel().trainer(env)
trainer.train()

And to load an already trained model:

env.download_model("training-job-name")
predictor = MyModel().predictor(env)
predictor.setup()

Happy local analyzing and debugging!