Detect and remove IDs disguised in text features #415
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
TuanNguyen27
requested review from
gerashegalov,
Jauntbox and
leahmcguire
as code owners
tovbinm
reviewed
Oct 11, 2019
| @@ -249,26 +257,11 @@ object FeatureDistribution { | |||
| nulls = nullCount, | |||
| summaryInfo = summaryInfo, | |||
| distribution = distribution, | |||
| moments = moments, | |||
There was a problem hiding this comment.
What about backwards compatibility?
tovbinm
reviewed
Oct 11, 2019
| @@ -142,7 +142,8 @@ case class RawFeatureFilterMetrics | |||
| scoringFillRate: Option[Double], | |||
| jsDivergence: Option[Double], | |||
| fillRateDiff: Option[Double], | |||
| fillRatioDiff: Option[Double] | |||
| fillRatioDiff: Option[Double], | |||
| trainingCardSize: Option[Int] | |||
There was a problem hiding this comment.
let's name it fully, i.e trainingCardinalitySize
tovbinm
reviewed
Oct 11, 2019
| @@ -0,0 +1,32 @@ | |||
| package com.salesforce.op.stages.impl.feature | |||
There was a problem hiding this comment.
License header is missing
tovbinm
reviewed
Oct 11, 2019
| import com.salesforce.op.features.types.{TextMap} | ||
| import com.salesforce.op.stages.base.unary.UnaryTransformer | ||
|
|
||
| class IdMapRemover( |
tovbinm
reviewed
Oct 11, 2019
| import com.salesforce.op.features.types.Text | ||
| import com.salesforce.op.stages.base.unary.UnaryTransformer | ||
|
|
||
| class IdRemover( |
tovbinm
reviewed
Oct 11, 2019
| minUniqueTokLen: Int, | ||
| uid: String = UID[IdRemover], | ||
| operationName: String = "IDremover" | ||
| ) extends UnaryTransformer[Text, Text] (operationName = operationName, uid = uid) { |
There was a problem hiding this comment.
no need to expose operationName on the IdRemover ctor args. instead do:
extends UnaryTransformer[Text, Text] (operationName = "IdRemover", uid = uid)
tovbinm
reviewed
Oct 11, 2019
| @@ -99,7 +95,7 @@ class FeatureDistributionTest extends FlatSpec with PassengerSparkFixtureTest wi | |||
| distribs(0).distribution.length shouldBe 100 | |||
| distribs(0).distribution.sum shouldBe 10000 | |||
| distribs.foreach(d => d.featureKey shouldBe d.name -> d.key) | |||
| distribs(0).moments.get.count shouldBe 10000 | |||
| // distribs(0).moments.get.count shouldBe 10000 | |||
tovbinm
reviewed
Oct 11, 2019
|
|
||
| @RunWith(classOf[JUnitRunner]) | ||
| class IdRemoverTest extends OpTransformerSpec[Text, IdRemover] { | ||
| val sample = Seq(Text("ball"), Text("stray"), Text("happy"), |
There was a problem hiding this comment.
please add newlines between the vals for readability
tovbinm
reviewed
Oct 11, 2019
| @@ -0,0 +1,17 @@ | |||
| package com.salesforce.op.stages.impl.feature | |||
tovbinm
reviewed
Oct 11, 2019
| @@ -56,30 +56,32 @@ import org.apache.spark.sql.SparkSession | |||
| * @param cabin cabin id string | |||
| * @param embarked location where passenger embarked | |||
| */ | |||
| case class Passenger | |||
| case class IDTextClassification | |||
There was a problem hiding this comment.
did this get here by mistake?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Related issues
We currently hash all text features whose cardinality is too high for pivoting. However, a text feature could contain all unique values (ID-like text, e.g SSN, zip code), which won't provide useful signal to the modeling via hashing. We would like Raw Feature Filter to exclude these features from feature engineering and model training.
Describe the proposed solution
We keep track of all the different lengths of the tokenized text. If a text feature has too few unique lengths, it's likely to not contain natural language text.
Describe alternatives you've considered
Cramer's V correlation between the distribution of hashed token, and a uniform distribution.
Count of k-th most frequent token in the text feature.
Portion of tokens that's covered by the top-k most frequent tokens.
Additional context