Skip to content

Latest commit

 

History

History
128 lines (95 loc) · 4.07 KB

README.md

File metadata and controls

128 lines (95 loc) · 4.07 KB
Raw

TransmogrifAI Command Line Interface (OP CLI)

Installation

  1. add to your bash profile: 
    export OP_HOME="/path/to/TransmogrifAI/"
alias op="java -cp $OP_HOME/cli/build/libs/\* com.salesforce.op.cli.CLI"
  2. run ./gradlew cli:shadowJar in your TransmogrifAI project directory
  3. now running op in terminal should run the most recently build shadow jar every time
  4. if you change the CLI, rerun ./gradlew cli:shadowJar

Usage

Run op --help to list the all the CLI options.

For instance, in order to generate a binary classification model with Titanic data one can do it simply by running:

op gen --input Passengers.csv --id passengerId --response survived --schema Passenger.avsc TitanicProj

If you have answers to questions in a file (see e.g. passengers.answers), and want to overwrite the old project, the command line looks like this:

op gen --input templates/simple/src/main/resources/PassengerData.csv --id passengerId --response survived --schema utils/src/main/avro/PassengerCSV.avsc --answers cli/passengers.answers TitanicProj10112017 --overwrite

Templates

Right now there is only one template, simple, defined in templates/simple. The template compiles with the rest of the project to make sure that there are no errors. Each file in the template will be copied over whenever a user runs the op gen command, and they will be run through a very tiny templating engine that allows substitutions. To exclude a file from templating (the best reason for this is if the file is binary), add it to the shouldCopy method for your template.

This templating engine has directives in comments that look like:

val (pred, raw, prob) = BinaryClassificationModelSelector() /* << PROBLEM_KIND */
    .setInput(label, checkedFeatures)
    .getOutput()

The templating engine reads comments like these and pulls the value to substitute (in this case PROBLEM_KIND) from the map of arguments created by the project template (for example, in here).

What this directive does is replace the previous "scala expression" with the value to substitute. A "scala expression", in this case, is a string of non-whitespace characters. Whitespace characters are included so long as there are unclosed brackets ([, (, or {).

To replace the next expression instead, do this:

object /* APP_NAME >> */ Simple extends OpApp with Features {

Examples

1.
val (pred, raw, prob) = BinaryClassificationModelSelector()   /* << PROBLEM_KIND */
    .setInput(label, checkedFeatures)
    .getOutput()

with Map("PROBLEM_KIND" -> "MY_VALUE") becomes:

val (pred, raw, prob) = MY_VALUE
    .setInput(label, checkedFeatures)
    .getOutput()
2.
object /* APP_NAME >> */ Simple extends OpApp with Features {

with Map("APP_NAME" -> "MY_VALUE") becomes:

object MY_VALUE extends OpApp with Features {
3.
trait Features extends Serializable /* FEATURES >> */ {

  val survived = FeatureBuilder.RealNN[Passenger]
    .extract(_.getSurvived.toDouble.toRealNN).asResponse

  val embarked = FeatureBuilder.MultiPickList[Passenger]
    .extract(d => Option(d.getEmbarked).toSet[String].toMultiPickList).asPredictor

}

with Map("FEATURES" -> "{\n\nhello world\n\n}") becomes:

trait Features extends Serializable {

hello world

}
4.
 val checkedFeatures = new SanityChecker()
    .setCheckSample(0.10)
    .setInput(label, featureVector) /* << NEW_INPUT */
    .getOutput()

with Map("NEW_INPUT" -> ".myMethod(blah)") becomes:

 val checkedFeatures = new SanityChecker()
    .setCheckSample(0.10)
    .myMethod(blah)
    .getOutput()

See also the relevant section on our docs site.