Getting Started with FastScore v1.10

This is a guide for installing and running FastScore. It contains instructions for first-time and novice users, as well as reference instructions for common tasks. This guide was last updated for v1.10 of FastScore.

If you need support or have questions, please email us: support@opendatagroup.com

Installing FastScore
Configuring and Starting FastScore
Using the FastScore Dashboard
Working with Models and Streams

Installing FastScore

This guide will walk you through installing and running Open Data Group’s FastScore microservices suite. The following instructions will generally assume that you’re working on a Linux machine. There are slight differences if you’re running FastScore on MacOS, which will be indicated by a special note. The differences if you’re running Windows have not yet been fully charted.

Prerequisites

The FastScore Microservices Suite is hosted on DockerHub (https://hub.docker.com/u/fastscore/). As such, one must first install Docker. For example, on Ubuntu Linux:

$ sudo apt-get install docker.io

It’s also useful (recommended but not mandatory) to have Docker Compose installed. Installation instructions can be found here: docs.docker.com/compose/install/.

On MacOS, Docker actually runs inside of a virtual machine (see Docker’s documentation here: https://docs.docker.com/machine/ ). In order to make sure all of the ports and IP addresses are handled correctly, you’ll need to run the commands from inside this virtual machine. To start the virtual machine and give it the name “default”, use the following command:
$ docker-machine create --driver=virtualbox default
This uses VirtualBox as the driver for the virtual machine. If you don’t have it already, you should download the VirtualBox client to manage the docker-machine. Among other things, this can be used to set up port forwarding for the virtual machine, which may be needed later. To switch to this environment for the default virtual machine, use the following command:
$ eval $(docker-machine env default)
The virtual machine’s IP address can be retrieved with the docker-machine ip command, e.g.,
$ docker-machine ip
192.168.99.100
This IP address should be used as the FastScore host machine IP address.

Once Docker has been installed, there are only a few steps needed to get FastScore running.

Configure FastScore using the FastScore CLI and configuration file.
Start the FastScore services, either manually, or via Docker Compose (recommended).
Connect to the FastScore Dashboard with your browser.
Write a FastScore configuration file.

Let’s go through each step carefully.

Installing the FastScore Command-Line Interface (CLI)

The FastScore CLI can be downloaded here:

Mac Linux Windows

For Mac (Darwin) and Linux make it executable: chmod +x fastscore

And finally add to $Path. For Mac users, copy to /usr/local/bin/.

The CLI is a statically-linked binary written in Go. We also have a Python implementation of the CLI available here.

Once you’ve installed the FastScore CLI, check that it works by executing the following command in your terminal. Also see FastScore Command Line Interface for more information on the CLI commands.

$ fastscore
NAME:
fastscore - Control FastScore from the command line.

USAGE:
fastscore [global options] command [command options] [arguments...]

VERSION:
v1.10

COMMANDS:
connect     Establish a FastScore Connection
use         Select the target engine instance
run         A shortcut for running a particular model and stream configuration
config      Configure the FastScore fleet
fleet       Examine status of the FastScore fleet
model       Manage analytic models
schema      Manage schemata
stream      Manage streams
sensor      Manage sensors
engine      Manage engines
attachment  Manage attachments
pneumo      Print pneumo messages
create      Create new FastScore services
destroy     Destroy existing FastScore services
conductor   Interact with the orchestrator
composer    Create and manage workflows
workflow    Create and manage workflows
monitor     Monitor an engine
login       Login to fastscore
debug       Watch debugging messages
stats       Get various statistics
help, h     Shows a list of commands or help for one command

GLOBAL OPTIONS:
--help, -h     show help
--version, -v  print the version

WEBSITE: http://www.opendatagroup.com
SUPPORT: support@opendatagroup.com

This displays a list of all of the FastScore CLI commands.

Start FastScore Microservices Suite with Docker Swarm (Recommended)

We will be referencing files from the Getting-Started github repository found here. Use the following command to clone the repo: git clone https://github.com/opendatagroup/Getting-Started.git

This set up will use Docker Swarm to bring up the FastScore containers.

Items in the repo:

README.md - descriptions of the files and instructions to deploy using FastScore
Makefile - automated setup and tear down of FastScore
docker-compose.yaml - contains a basic collection of FastScore containers to deploy models
config.yaml - configures the FastScore fleet
setup.sh - referenced in the Makefile to set up FastScore
load.sh - referenced in the Makefile to load in play models, attachments, schemas, and streams
library/ - contains models, attachments, schemas, and streams
data/ - contains sample data

From the directory with the Makefile, to quickly start FastScore:

make deploy

To quickly tear down FastScore:

make stop

First, let’s look at the containers described in the docker-compose.yaml file that are now up and running.

View all Docker containers that are running with the docker ps command. The output should look something like this:

CONTAINER ID        IMAGE                              COMMAND                  CREATED             STATUS              PORTS                                    NAMES
14bfaee9100e        fastscore/model-manage-mysql:1.9   "sh -c 'cd /var/lib/…"   About an hour ago   Up About an hour                                             fs-vanilla_database.1.vb7bticmnpauo3hz6bor4r8qx
ada9681679e8        fastscore/engine:1.10               "/fastscore/startup.…"   About an hour ago   Up About an hour                                             fs-vanilla_engine-2.1.ubonr83uuaqw2en4yk3lsc9w0
880b6882742d        fastscore/engine:1.10               "/fastscore/startup.…"   About an hour ago   Up About an hour                                             fs-vanilla_engine-1.1.or1ocif21hfoej8xxfo5qykbm
cb67e0090cf2        fastscore/connect:1.10              "bin/connect"            About an hour ago   Up About an hour                                             fs-vanilla_connect.1.wxswv0g8hojps2f2c0i5lc96l
318eedda837a        fastscore/dashboard:1.10            "npm --no-update-not…"   About an hour ago   Up About an hour                                             fs-vanilla_dashboard.1.v7ia7we8qrwpyeawkk7jzg3se
b89be509f74d        fastscore/kafka:1.9                "/entry.sh"              About an hour ago   Up About an hour    2181/tcp, 2888/tcp, 3888/tcp, 9092/tcp   fs-vanilla_kafka.1.w1tc6grz7iil56x2b8pcdbjal
84b4f75fccda        fastscore/model-manage:1.9         "bin/model_manage"       About an hour ago   Up About an hour                                             fs-vanilla_model-manage.1.o750isigzx7v17kwddn17nmzl

Configuring and Starting FastScore

Next, let’s take a look at the setup.sh file that is referenced in the Makefile.

fastscore connect https://localhost:8000
fastscore config set config.yaml
fastscore fleet -wait

FastScore will ask you to connect to the host it is running on and the port of the dashboard. In this example we are using localhost and the dashboard is listening on port 8000.

FastScore’s microservices architecture requires each microservice component to communicate with other components. These communications are managed by the Connect microservice. In order for Connect to connect, it has to be given information about the other microservices components in a configuration file. A sample configuration file is shown below:

fastscore:
  fleet:
    - api: model-manage
      host: model-manage
      port: 8002
    - api: engine
      host: engine-1
      port: 8003
    - api: engine
      host: engine-2
      port: 8003

  db:
    type: mysql
    host: database
    port: 3306
    username: root
    password: root

  pneumo:
    type: REST
    #type: kafka
    #bootstrap:
    #  - kafka:9092
    #topic: notify

Configuration files are written in YAML. The configuration file above specifies the host machines and ports for the Model Manage container, the MySQL database container used by Model Manage, and two Engine containers, all hosted on the same machine. Additionally, Pneumo, an asynchronous notification library used by FastScore, is configured to communicate via Kafka or REST.

We can then check the status of our containers using the fleet command:

$ fastscore fleet -wait
Name            API           Health
--------------  ------------  --------
engine-1        engine        ok
engine-2        engine        ok
model-manage-1  model-manage  ok

Finally, the load.sh file will load in some models, attachments, schemas, and streams into FastScore so you can start deploying! We are loading these into the MySql backing of Model Manage for this example. Alternatively, you can connect Model Manage to your own github repository that already has models, attachments, streams, and schemas in it! Now we’re ready to start scoring.

Using the FastScore Dashboard

FastScore’s Dashboard provides a convenient user interface for reviewing engine status and managing models, schemas, sensors, and streams. However, as compared to the FastScore CLI, it requires a few additional setup steps to get things running.

First, if you are not running FastScore on your local machine (for example, if you have FastScore running on a cloud service platform), you will need to allow incoming and outgoing traffic on port 8000 (used by the FastScore Dashboard). You will also need to have configured FastScore as described in the previous section.

To access the Dashboard, take your browser to the FastScore host machine at port 8000 using the HTTPS protocol (FastScore does not allow HTTP). If all goes well , you will be greeted by this screen:

Home Screen On the left-hand side of the Dashboard are four sections: engine-1, engine-2, model-manage-1, Connect. These correspond to the Engine microservices, the Model Manage microservice, and the Connect microservice. The green dots on the engines and model manage indicate that they are currently running correctly. If you have configured additional engine containers, they will also appear on the side.

If, instead, you get an “Application Initialization Error,” check your configuration file for any errors, and verify that you have followed all of the FastScore CLI configuration steps. If the fastscore fleet command shows both Model Manage and your Engine containers working properly, then the problem most likely has to do with Dashboard’s proxy service or your host machine’s network traffic settings.

Working with Models and Streams

FastScore leverages a set of lightweight, but powerful abstractions to allow the deployment, monitoring, and management of models across frameworks, languages, platforms, and environments. These abstractions fundamentally allow models to read in records from a data stream, score them, and output that score to another data stream. As such, running any model consists of three steps:

Loading models, streams, schemas, and sensors
Setting Engine parameters
Running the model

Creating and Loading Assets into FastScore Model Manage

Version 1.10 of FastScore supports models in Python, R, Java, MATLAB, PFA, PrettyPFA, H2O, PMML, and C formats. Some setup steps differ slightly between Python/R models and PFA, Java, MATLAB, or C models. As a model interchange format, PFA can provide some benefits in performance, scalability, and security relative to R and Python. PrettyPFA is a human-readable equivalent to PFA. However, as the majority of users will be more familiar with R and Python, we focus on these two languages in this section.

Loading Assets

The FastScore CLI allows a user to load models directly from the command line. The list of models currently loaded in FastScore can be viewed using the model list command:

$ fastscore model list
Name    Type
------  ------
MyModel Python

Models can be added with model add <name> <file>, and removed with model remove <name>. Additionally, the fastscore model show <name> command will display the named model.

Models via the Dashboard

The Dashboard provides functionality to add and manage models. To upload a model, under the Models tab, select the “Upload model” button, and choose a model from your local machine. Alternatively, “select model”, depicted below, allows you to select an existing model from the model manager by name.

Model Load

Additionally, models can be added, removed, inspected, and edited from the Models tab under Model Manage:

Model Load The screenshot above shows the model manager tab, and an existing model in FastScore. Models can be removed, saved, created, uploaded, or downloaded from this view. Note that after creating or modifying a model in this view, it must still be selected for use from the Engine tab.

Models in Python and R

All models are added to FastScore and executed using the same CLI commands, namely:

fastscore model add <modelname> <path/to/model.extension>

Note that, in order to determine whether a model is Python or R, Engine requires that it have an appropriate file extension (.py for Python, .R for R, .pfa for PFA, .ppfa for PrettyPFA, etc.). Also, in order to score a Python/R model, there are certain constraints on the form the model must take.

FastScore includes both a Python2 and Python3 model runner. By default, .py files are interpreted as Python2 models—to load a Python3 model, use the file extension .py3, or the flag -type:python3 option with fastscore model add:

fastscore model add -type:python3 my_py3_model path/to/model.py

Python Models

There are two ways to write models to run in FastScore for Python. Style 1New in the 1.10 release, we now support an additional style of conformance to make it easier for Data Scientist to get their models ready for deployment for even more classes of models. This style is geared for models that follow a classic pattern of: (a) read data in (b) perform a prediction, and (c) write the output to a consumer. This new approach uses slot objects to control the scoring behavior of the model as shown in this example below. For a full guide, refer to this how-to and example.

# fastscore.action: unused

import cPickle
import FeatureTransformer
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.pipeline import Pipeline

from fastscore.io import Slot

with open("gbmFit.pkl", "rb") as f:
    gbm = cPickle.load(f)
for df in Slot(0):
    Slot(1).write(gbm.predict(df))

Style 2: An alternative approach that is well suited to a variety of REST-based paradigms is the existing ‘call-back’ style, which leverages begin() and action() functions controlling the prediction behavior. All Python models must declare a one-argument action() function. This function will be called everytime data is being score. The minimal example of a Python model is the following:

# fastscore.input: input-schema
# fastscore.output: output-schema

def action(datum):
    yield 0

This extremely simple model produces a 0 for every input.

In addition, we can declare a begin() function, which is called when the model is initallly deployed into the Engine and awaiting data to score.

A slightly more sophisticated example of a Python model is the following:

# fastscore.input: input-schema
# fastscore.output: output-schema

import cPickle as pickle

def begin(): # perform any initialization needed here
        global myObject
    myObject = pickle.load(open('object.pkl'))
    pass # or do something with the unpickled object

def action(datum): # datum is expected to be of the form '{"x":5, "y":6}'
    record = datum
    x = record['x']
    y = record['y']
    yield x + y

def end():
    pass

This model returns the sum of two numbers. Note that we are able to import Python’s standard modules, such as the pickle module. Non-default packages can also be added by using #fastscore.module-attached: <module-nam> once installed in the Engine via the Dockerfile. Custom classes and packages can be loaded using attachments, as described in the Gradient Boosting Regressor tutorial.

R Models

R models feature much of the same functionality as Python models, as well as the same constraint: the user must define an action function to perform the actual scoring. For example, the analogous model to the Python model above is

# fastscore.input: input-schema
# fastscore.output: output-schema

# Sample input: {"x":5.0, "y":6.0}
action <- function(datum) {
  x <- datum$x
  y <- datum$y
  emit(x + y)
}

Input and Output Schema

FastScore enforces strong typing on both the inputs and outputs of its models using AVRO schema. For R and Python models, this typing is enforced by specifying schema names in a smart comment at the top of the model file:

# fastscore.input: array-double
# fastscore.output: double

Python and R models must specify schemas for their inputs and outputs. PrettyPFA and PFA models already contain the input and output schema as part of the model definition, so they do not require a schema attachment.

For example, a model that expects to receive records of two doubles as inputs might have the following schema:

{
  "name": "Input",
  "type": "record",
  "fields" : [
    {"name": "x", "type": "double"},
    {"name": "y", "type": "double"}
  ]
}

The model might then produce a stream of doubles as its output:

{
  "name": "Output",
  "type": "double"
}

Input and output schema must be uploaded separately to FastScore. To upload the schema to FastScore with the CLI, use the following commands:

fastscore schema add input input.avsc
fastscore schema add output output.avsc

The 1.10 release also includes new funcitonality to automatically generate the schema from a sample data file using the fastscore schema infer <data-file>.

Schemas can also be managed from within the Dashboard, using the Model Manage view.

Input and Output Streams

Before a model can be run, it has to have some data to run on. Input and output streams are used to supply the incoming data to the model, and to return the corresponding scores. FastScore supports a wide variety of stream transports including: file, Kafka, Authenticated Kafka, Executable, HTTP, TCP, UDP, ODBC, debug, and console streams. All of these types are configured using a Stream Descriptor file.

Stream Descriptors are small JSON files containing information about the stream. An example of a Stream Descriptor for a Kafka stream is displayed below:

{
  "Description": "read Avro-typed JSON records from a Kafka stream",
  "Transport": {
    "Type": "kafka",
    "BootstrapServers": ["127.0.0.1:9092"],
    "Topic": "data-feed-1",
    "Partition": 0
  },
  "Encoding": "json",
  "Schema": { type: "record", ... }
}

Stream descriptors are documented in more detail on the stream descriptor page. The easiest type of stream to use is a file stream, which reads or writes records directly from/to a file inside of the FastScore engine container. Here is an example of such a stream:

{
  "Description": "read input from the specified file",
  "Loop": false,
  "Transport": {
    "Type": "file",
    "Path": "/root/data/neural_net_input.jsons"
  },
  "Envelope": "delimited",
  "Encoding": "json",
  "Schema": {"type": "array", "items": "double"}
}

This file stream expects each line of the neural_net_input.jsons file to be a vector of doubles, encoded as a JSON object, and delimitated by newlines. The file is located in the /root/data/ directory of the engine container. The "Loop": false line tells FastScore to stop reading the file after reaching the end of the file, as opposed to looping over the lines in the file.

Streams via FastScore CLI

The FastScore CLI can be used to configure data streams. The stream list command displays a list of existing streams:

$ fastscore stream list
demo-1
demo-2

By default, two demo file streams are included in FastScore. The demo-1 data set consists of random numbers. The demo-2 dataset consists of lists of JSONS with the following AVRO schema:

{
  "type":"array",
  "items": {
    "type": "record",
    "fields": [
      {"name":"x", "type":"double"},
      {"name":"y", "type":"string"}]
  }
}

These demo streams can be used to test whether or not a simple model is working correctly. Additional streams can be added using the fastscore stream add <stream-name> <stream-descriptor-file> command. Existing streams can be sampled (displaying the most recent items of the stream) with fastscore stream sample <stream-name>.

For filestreams, it is easiest to manage container input and output by linking a directory on the host machine to the engine container. This can be done in the Docker-Compose file by modifying the engine service to the following:

[...]

  engine-1:
    image: fastscore/engine:1.10
    network_mode: "host"
    environment:
      CONNECT_PREFIX: https://127.0.0.1:8001
    volumes:                           # new volume section
      - ./data:/root/data


[...]

This will link the ./data directory on the host machine to the /root/data directory of the engine container. A filestream from the file “mydata.jsons” located in data on the host machine can then be accessed by FastScore using the stream descriptor

{
  "Loop": false,
  "Transport": {
    "Type": "file",
    "Path": "/root/data/mydata.jsons"
  },
  "Envelope": "delimited",
  "Encoding": "json",
  "Schema": [...]
}

A similar stream descriptor can be used for the output stream to write the output scores to a file in the same directory.

When using Docker volume linking to link a directory on the host machine to the Engine instance, Docker must have privileges to read and write from the specified directory. Additionally, the directory on the container must be chosen carefully, as its contents will be overwritten with the contents of the corresponding host directory upon linking. /root/data is safe (as it only contains the demo datafiles), but other directories on the container (e.g., /usr) may not be.

Streams via the Dashboard

Analogously to models, streams can also be managed from the Dashboard. Selecting the “Streams” tab under Model Manage displays the following view:

Streams On the left, existing Stream Descriptors are displayed. New Stream Descriptors can be added and existing ones edited from this view. The example above displays a simple file stream, which will load the schemaInputArrayDouble.jsons file located in the /tmp/data/ directory of the Engine Docker container.

Engine Parameters

Engine parameters, such as the number of Engine instances currently running, as well as information about the model, are displayed on the Dashboard Engine tab.

Running a Model in FastScore

When using the Dashboard, models will begin scoring as soon as both the model and input/output streams are set from the Engine tab, and no further action from the user is required. Various statistics about performance and throughput are displayed on the Engine tab.

To run a model using the FastScore CLI, use the fastscore run sequence of commands:

fastscore use engine-1 selects the Engine where we will deploy the model.
fastscore engine reset refreshes the Engine to make it ready for deployment.
fastscore run <model-name> <input-stream-name> <output-stream-name> runs the model named <model-name> with the specified input and output streams.
fastscore run <model-name> rest: rest: You can also use rest: instead of the stream name to deploy the model with a generated REST endpoint.
fastscore model interact allows you to send sample data and recieve the output back from the CLI.
fastscore engine stop halts the currently running model.
fastscore engine inspect displays the status of the Engine. If the Engine is in the ERROR state, view the Docker Logs to view the error message.

Some of the statistics displayed by the fastscore job statistics command, such as memory usage, are also shown on the Dashboard.

This concludes the FastScore Getting Started guide. Now that you’re familar with FastScore, check out our guide on Conforming and Deploying a Model for a detailed walkthrough to help you get your model ready for deployment in FastScore!

Additional FastScore API documentation is available at https://opendatagroup.github.io/Reference/FastScore%20API/. Happy scoring!