edfernandez-NU/Docker-Spark-MachineLearning

A microservice for serving Machine Learning models created in Spark

Containerized Model Serving using Databricks Model Export Library and Docker

This demo only works for zip code in the Seattle metro area. In order to run this with your own model, replace the "pipeline" under the resources folder with your own.

(Option 1) For running in interactive mode

$ docker run -it vedantja/realestate_price_predictor

You should see the following output:

Enter information about the house below.
> number of bedrooms: 5 
> number of bathrooms: 3 
> sq footage of living area: 1000
> sq footage of the lot: 2000
> zipcode: 98039
> condition: 5

Mar 13, 2018 8:10:38 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 PipelineModel model with 0 rows from pipeline/data.
Mar 13, 2018 8:10:38 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 feature.StringIndexerModel model with 1 rows from pipeline/stages/0_StringIndexer_453bb914baee6462c1db/data.
Mar 13, 2018 8:10:38 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 feature.VectorAssembler model with 0 rows from pipeline/stages/1_VectorAssembler_4b21b95f4a78010ae379/data.
Mar 13, 2018 8:10:38 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 regression.DecisionTreeRegressionModel model with 63 rows from pipeline/stages/2_DecisionTreeRegressor_4dd187cc681dd86b9635/data.

And the prediction will be output like this:

{"prediction":571955.1501706485} 

(Option 2a) For running as a Microservice with the housing model

$ docker run -p 15000:5000 vedantja/model_microservice_example

You should see the following output:

vjain:docker vedantjain$ docker run -p 15000:5000 vedantja/model_microservice_example
 * Restarting with stat
 * Debugger is active!
 * Debugger PIN: 556-327-343
 * Running on http://0.0.0.0:5000/ (Press CTRL+C to quit)

The application is running on port 5000. So open another terminal window and submit the following command:

$ curl -H 'Content-Type:application/json' -d '{"bedrooms": 5, "bathrooms": 3, "sqft_living": 1000, "sqft_lot": 2000, "zipcode": "98039", "condition": 5}' localhost:5000

(Option 2b) For running with your own model in the microservice

You can use this to get predictions from your own model. Following are the prerequisites:

1. Model/Pipeline should have been exported using Databricks Model Export library
2. The model exprt should be stored on an S3 bucket and in zip format
3. Access key and secret access for an IAM user with ability to read from S3
4. Name of the S3 bucket and full path of the file
5. Git clone of this repository

See example below:

1. Clone the repo

$ git clone https://github.com/vedantja/Docker-Spark-MachineLearning.git

2. Go to the folder

$ cd Docker-Spark-MachineLearning/byom_microservice

3. Build the docker image

$ docker build -t byom_microservice .

4. Start a container

$ docker run -e access_key="xxxxxxxxxxxxxxx" -e secret_access="xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" -e bucket_name="xxxxxxxxxxxxx" -e zip_file_path="/path/to/zipfile" -p 15000:5000 byom_microservice

You should see the following output:

vjain:docker vedantjain$ docker run -p 15000:5000 vedantja/model_microservice_example
 * Restarting with stat
 * Debugger is active!
 * Debugger PIN: 556-327-343
 * Running on http://0.0.0.0:5000/ (Press CTRL+C to quit)

The application is running on port 5000. So open another terminal window and submit the following command:

$ curl -H 'Content-Type:application/json' -d '{"bedrooms": 5, "bathrooms": 3, "sqft_living": 1000, "sqft_lot": 2000, "zipcode": "98039", "condition": 5}' localhost:5000<

You will see the following output in the new window:

{"prediction":571955.1501706485}

The output in the docker console should look something like this:

--------------------------------------------------------------------------------
INFO in app [./app.py:13]:
POST request received from: 172.17.0.1
--------------------------------------------------------------------------------
Mar 15, 2018 7:48:27 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 PipelineModel model with 0 rows from pipeline/data.
Mar 15, 2018 7:48:27 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 feature.StringIndexerModel model with 1 rows from pipeline/stages/0_StringIndexer_453bb914baee6462c1db/data.
Mar 15, 2018 7:48:27 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 feature.VectorAssembler model with 0 rows from pipeline/stages/1_VectorAssembler_4b21b95f4a78010ae379/data.
Mar 15, 2018 7:48:27 PM com.databricks.ml.local.util.Logging$class logInfo
INFO: Importing version 0.1.1 regression.DecisionTreeRegressionModel model with 63 rows from pipeline/stages/2_DecisionTreeRegressor_4dd187cc681dd86b9635/data.
172.17.0.1 - - [15/Mar/2018 19:48:27] "POST / HTTP/1.1" 200 -