H2O Flow
H2O Flow is a web-based interactive computational environment where you can combine code execution, text, mathematics, plots and rich media to build machine learning workflows.
Think of Flow as a hybrid GUI + REPL + storytelling environment for exploratory data analysis and machine learning, with async, re-scriptable record/replay capabilities. Flow sandboxes and evals user-Javascript in the browser via static analysis and tree-rewriting. Flow is written in CoffeeScript, with a veritable heap of little embedded DSLs for reactive dataflow programming, markup generation, lazy evaluation and multicast signals/slots.
Docs
There is a nice user guide for H2O Flow housed over in the h2o-3 repo.
Development Setup
It is recommended that you clone h2o-3 and h2o-flow in the same parent directory.
If you have not already, follow these instructions to set up your preferred IDE environment for h2o-3 development.
-
First build H2O-3
cd h2o-3 && ./gradlew build -x test
(in h2o-E) -
Install npm dependencies for h2o-flow
npm i
(in h2o-flow)
Developing with live reload
-
Start H2O-3 with CORS checks disabled
java -Dsys.ai.h2o.disable.cors=true -jar build/h2o.jar
(in h2o-3) -
Start webpack dev-server
npm run start
(in h2o-flow)
This will open a browser window with auto-refreshing dev server.
Development within h2o-3 instance
-
Run
make
command. This will copy the build resources into the neighbouring h2o-3 directory. -
Start h2o-3 from IDE without running gradle (which would write over your local flow build)
Testing a new Flow Feature with Sparkling Water
Flow can also be used with Sparkling Water
Follow this guide develop and test new Sparkling Water features in Flow.
adapted from the comments on this PR #13
copy built js files from one place to another
in the h2o-3
directory run:
cp h2o-web/src/main/resources/www/flow/js/* h2o-web/lib/h2o-flow/build/js/
build h2o-3
in the h2o-3
directory run:
./gradlew publishToMavenLocal -x test
build sparkling water
in sparkling-water
directory run:
./gradlew clean build -x test -x integTest
open the Sparkling Water Shell
in sparkling-water
directory run:
bin/sparkling-shell
in the sparkling water shell
at the scala>
prompt run:
import org.apache.spark.h2o._
H2OContext.getOrCreate(sc)
now open Flow at the IP address specified
in the sparkling water shell
now test your changes in Flow