Adjust bin values for text features in RFF #99
There are no files selected for viewing
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
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
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -31,10 +31,10 @@ | |
| package com.salesforce.op.filters | ||
|
|
||
| import com.salesforce.op.features.FeatureDistributionLike | ||
| import com.salesforce.op.stages.impl.feature.{HashAlgorithm, Inclusion, NumericBucketizer} | ||
| import com.twitter.algebird.Monoid._ | ||
| import com.twitter.algebird.Operators._ | ||
| import com.twitter.algebird.Semigroup | ||
| import org.apache.spark.mllib.feature.HashingTF | ||
|
|
||
| /** | ||
|
|
@@ -44,7 +44,7 @@ import org.apache.spark.mllib.feature.HashingTF | |
| * @param key map key associated with distribution (when the feature is a map) | ||
| * @param count total count of feature seen | ||
| * @param nulls number of empties seen in feature | ||
| * @param distribution binned counts of feature values (hashed for strings, evenly spaced bins for numerics) | ||
| * @param summaryInfo either min and max number of tokens for text data, | ||
| * or splits used for bins for numeric data | ||
| */ | ||
|
|
@@ -125,8 +125,8 @@ case class FeatureDistribution | |
| def jsDivergence(fd: FeatureDistribution): Double = { | ||
| checkMatch(fd) | ||
| val combinedCounts = distribution.zip(fd.distribution).filterNot{ case (a, b) => a == 0.0 && b == 0.0 } | ||
| val (thisCount, thatCount) = | ||
| combinedCounts.fold[(Double, Double)]((0.0, 0.0)){ case ((a1, b1), (a2, b2)) => (a1 + a2, b1 + b2) } | ||
| val probs = combinedCounts.map{ case (a, b) => a / thisCount -> b / thatCount } | ||
| val meanProb = probs.map{ case (a, b) => (a + b) / 2} | ||
| def log2(x: Double) = math.log10(x) / math.log10(2.0) | ||
|
|
@@ -154,19 +154,19 @@ private[op] object FeatureDistribution { | |
| * @param summary feature summary | ||
| * @param value optional processed sequence | ||
| * @param bins number of histogram bins | ||
| * @param textBinsFormula formula to compute the text features bin size | ||
| * @return feature distribution given the provided information | ||
| */ | ||
| def apply( | ||
| featureKey: FeatureKey, | ||
| summary: Summary, | ||
| value: Option[ProcessedSeq], | ||
| bins: Int, | ||
| textBinsFormula: (Summary, Int) => Int | ||
| ): FeatureDistribution = { | ||
| val (nullCount, (summaryInfo, distribution)): (Int, (Array[Double], Array[Double])) = | ||
| value.map(seq => 0 -> histValues(seq, summary, bins, textBinsFormula)) | ||
| .getOrElse(1 -> (Array(summary.min, summary.max, summary.sum, summary.count) -> new Array[Double](bins))) | ||
|
|
||
| FeatureDistribution( | ||
| name = featureKey._1, | ||
|
|
@@ -179,40 +179,46 @@ private[op] object FeatureDistribution { | |
|
|
||
| /** | ||
| * Function to put data into histogram of counts | ||
| * | ||
| * @param values values to bin | ||
| * @param summary summary info for feature (max, min, etc) | ||
| * @param bins number of bins to produce | ||
| * @param textBinsFormula formula to compute the text features bin size | ||
| * @return the bin information and the binned counts | ||
| */ | ||
| private def histValues( | ||
| values: ProcessedSeq, | ||
| summary: Summary, | ||
| bins: Int, | ||
| textBinsFormula: (Summary, Int) => Int | ||
| ): (Array[Double], Array[Double]) = values match { | ||
| case Left(seq) => | ||
| val numBins = textBinsFormula(summary, bins) | ||
|
There was a problem hiding this comment. why would the formula need the bins? There was a problem hiding this comment. Right now we set is as There was a problem hiding this comment. what would you recommend then? @leahmcguire There was a problem hiding this comment. bins is the default number (e.g. 100) there might be cases that we will just want to fall back to the defaults |
||
| // TODO: creating too many hasher instances may cause problem, efficiency, garbage collection etc | ||
| val hasher = | ||
| new HashingTF(numFeatures = numBins).setBinary(false) | ||
| .setHashAlgorithm(HashAlgorithm.MurMur3.entryName.toLowerCase) | ||
| Array(summary.min, summary.max, summary.sum, summary.count) -> hasher.transform(seq).toArray | ||
|
|
||
| case Right(seq) => // TODO use kernel fit instead of histogram | ||
| if (summary == Summary.empty) { | ||
| Array(summary.min, summary.max) -> seq.toArray // the seq will always be empty in this case | ||
| } else if (summary.min < summary.max) { | ||
| // total number of bins includes one for edge and one for other | ||
| val step = (summary.max - summary.min) / (bins - 2.0) | ||
| val splits = (0 until bins).map(b => summary.min + step * b).toArray | ||
| val binned = seq.map { v => | ||
| NumericBucketizer.bucketize( | ||
| splits = splits, trackNulls = false, trackInvalid = true, | ||
| splitInclusion = Inclusion.Left, input = Option(v) | ||
| ).toArray | ||
| } | ||
| val hist = binned.fold(new Array[Double](bins))(_ + _) | ||
| splits -> hist | ||
| } else { | ||
| val same = seq.map(v => if (v == summary.max) 1.0 else 0.0).sum | ||
| val other = seq.map(v => if (v != summary.max) 1.0 else 0.0).sum | ||
| Array(summary.min, summary.max, summary.sum, summary.count) -> Array(same, other) | ||
| } | ||
| } | ||
| } | ||
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
Oops, something went wrong.
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.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why the int?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
it's the bins.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah but that is not what we want the formula to look like long term
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
so what is your suggestion on the inputs / outputs then?