Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

fix!: statistical functions should return null when provided a vector of only null values #5606

Merged
merged 19 commits into from
Jun 19, 2024
Merged

fix!: statistical functions should return null when provided a vector of only null values #5606

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
076e286
Updated Numeric calculations and tests, corrected median and percenti…
lbooker42 Jun 11, 2024
dd9f507
Correct Numeric avg() function and tests.
lbooker42 Jun 12, 2024
5bf465e
UpdateBy corrections for all-NULL vectors
lbooker42 Jun 12, 2024
ad2675d
Agg corrections for avg,var,std
lbooker42 Jun 12, 2024
50667f2
Agg test changes to reflect new output.
lbooker42 Jun 13, 2024
ff6f010
Spotless
lbooker42 Jun 13, 2024
9ff3cf6
Tiny change for correctness.
lbooker42 Jun 13, 2024
8753267
Addressed PR comments.
lbooker42 Jun 17, 2024
4481f79
Addressed another round of PR comments.
lbooker42 Jun 17, 2024
9422488
More comments addressed.
lbooker42 Jun 17, 2024
5750546
Bug fix to pass OOB.
lbooker42 Jun 18, 2024
b06a063
Merge branch 'main' into lab-numeric-harmony
lbooker42 Jun 18, 2024
35c03d1
Spotless
lbooker42 Jun 18, 2024
0f150ac
Undo test commenting.
lbooker42 Jun 18, 2024
4e01d9d
Better testing and coverage, fix to wvar() and misc cleanup.
lbooker42 Jun 18, 2024
2ea6418
Test correction.
lbooker42 Jun 18, 2024
e00e49a
Removed dead code.
lbooker42 Jun 18, 2024
6bf8bb9
Update engine/function/src/templates/Numeric.ftl
lbooker42 Jun 18, 2024
cf1f22e
Update engine/function/src/templates/Numeric.ftl
lbooker42 Jun 18, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
134 changes: 94 additions & 40 deletions engine/function/src/templates/Numeric.ftl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1268,38 +1268,63 @@ public class Numeric {

if (n == 0) {
return NULL_DOUBLE;
} else {
// NULL values sorted to beginning of the array.
${pt.primitive}[] copy = sort(values.toArray());
}

${pt.primitive}[] sorted = values.copyToArray();
Arrays.sort(sorted);

// Determine if there are any NULL in the array.
int nullCount = 0;
for (int i = 0; i < n; i++) {
if (isNull(copy[i])) {
int nullStart = -1;
int nullCount = 0;

<#if pt.valueType.isFloat >
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
for (int i = 0; i < n; i++) {
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
final ${pt.primitive} val = sorted[i];
if (isNaN(val)) {
return Double.NaN; // Any NaN will pollute the result.
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}
if (nullStart == -1) {
if (isNull(val)) {
nullStart = i;
nullCount++;
} else {
break;
}
} else if (isNull(val)) {
nullCount++;
}

if (nullCount == 0) {
// No NULL, so we can just compute the median and return.
if (n % 2 == 0)
return 0.5 * (copy[n / 2 - 1] + copy[n / 2]);
else return copy[n / 2];
} else if (nullCount < n) {
// Some NULL, reduce the count and compute the median of the non-null values.
n -= nullCount;
if (n % 2 == 0) {
int index = n / 2;
return 0.5 * (copy[n / 2 - 1 + nullCount] + copy[n / 2 + nullCount]);
}
<#else>
for (int i = 0; i < n; i++) {
final ${pt.primitive} val = sorted[i];
if (nullStart == -1) {
if (isNull(val)) {
nullStart = i;
nullCount++;
}
else return copy[n / 2 + nullCount];
} else if (isNull(val)) {
nullCount++;
} else {
// All values are NULL.
return NULL_DOUBLE;
break; // no more NULL possible
}
}
</#if>

if (nullCount == n) {
return NULL_DOUBLE;
} else if (nullCount > 0) {
if (nullStart > 0) {
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
// Move the pre-NULL data so we have a contiguous block of non-NULL values.
System.arraycopy(sorted, 0, sorted, nullCount, nullStart);
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}
n -= nullCount;
if (n % 2 == 0) {
int index = n / 2;
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
return 0.5 * (sorted[n / 2 - 1 + nullCount] + sorted[n / 2 + nullCount]);
}
else return sorted[n / 2 + nullCount];
} else {
if (n % 2 == 0)
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
return 0.5 * (sorted[n / 2 - 1] + sorted[n / 2]);
else return sorted[n / 2];
}
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}

/**
Expand Down Expand Up @@ -1334,31 +1359,60 @@ public class Numeric {
}

int n = values.intSize("percentile");
// NULL values sorted to beginning of the array.
${pt.primitive}[] copy = sort(values.toArray());

// Determine if there are any NULL in the array.
if (n == 0) {
return ${pt.null};
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}

${pt.primitive}[] sorted = values.copyToArray();
Arrays.sort(sorted);

int nullStart = -1;
int nullCount = 0;

<#if pt.valueType.isFloat >
for (int i = 0; i < n; i++) {
if (isNull(copy[i])) {
final ${pt.primitive} val = sorted[i];
if (isNaN(val)) {
return ${pt.boxed}.NaN; // Any NaN will pollute the result.
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}
if (nullStart == -1) {
if (isNull(val)) {
nullStart = i;
nullCount++;
}
} else if (isNull(val)) {
nullCount++;
}
}
<#else>
for (int i = 0; i < n; i++) {
final ${pt.primitive} val = sorted[i];
if (nullStart == -1) {
if (isNull(val)) {
nullStart = i;
nullCount++;
}
} else if (isNull(val)) {
nullCount++;
} else {
break;
break; // no more NULL possible
}
}
</#if>

if (nullCount == 0) {
// No NULL, so we can just compute the index and return.
int idx = (int) Math.round(percentile * (n - 1));
return copy[idx];
} else if (nullCount < n) {
// Some NULL, reduce the count and compute the median of the non-null values.
n -= nullCount;
int idx = (int) Math.round(percentile * (n - 1));
return copy[idx + nullCount];
} else {
// All values are NULL.
if (nullCount == n) {
return ${pt.null};
} else if (nullCount > 0) {
if (nullStart > 0) {
// Move the pre-NULL data so we have a contiguous block of non-NULL values.
System.arraycopy(sorted, 0, sorted, nullCount, nullStart);
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}
int idx = (int) Math.round(percentile * (n - nullCount - 1));
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
return sorted[idx + nullCount];
} else {
int idx = (int) Math.round(percentile * (n - 1));
return sorted[idx];
}
}

Expand Down
16 changes: 16 additions & 0 deletions engine/function/src/templates/TestNumeric.ftl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1150,6 +1150,12 @@ public class TestNumeric extends BaseArrayTestCase {
assertEquals(3.0, median(new ${pt.boxed}[]{(${pt.primitive})4,(${pt.primitive})2,(${pt.primitive})3,${pt.null},${pt.null}}));
assertEquals(3.5, median(new ${pt.boxed}[]{(${pt.primitive})4,(${pt.primitive})2,(${pt.primitive})3,(${pt.primitive})5,${pt.null},${pt.null}}));

<#if pt.valueType.isFloat >
assertEquals(Double.NaN, median(new ${pt.primitive}[]{4,2,3, ${pt.boxed}.NaN}));
assertEquals(3.0, median(new ${pt.primitive}[]{4,2,3, ${pt.boxed}.POSITIVE_INFINITY, ${pt.null}, ${pt.null}, ${pt.boxed}.NEGATIVE_INFINITY}));
assertEquals(3.5, median(new ${pt.primitive}[]{4,2,3,5, ${pt.boxed}.POSITIVE_INFINITY, ${pt.null}, ${pt.null}, ${pt.boxed}.NEGATIVE_INFINITY}));
</#if>

// check that functions can be resolved with varargs
assertEquals(3.0, median((${pt.primitive})4, (${pt.primitive})2, (${pt.primitive})3));
}
Expand Down Expand Up @@ -1188,6 +1194,16 @@ public class TestNumeric extends BaseArrayTestCase {
assertEquals((${pt.primitive})2, percentile(0.00, new ${pt.primitive}[]{4,2,3,${pt.null}}));
assertEquals((${pt.primitive})3, percentile(0.50, new ${pt.primitive}[]{4,2,3,${pt.null},${pt.null}}));
assertEquals((${pt.primitive})4, percentile(1.0, new ${pt.primitive}[]{4,2,3,${pt.null},${pt.null},${pt.null}}));

<#if pt.valueType.isFloat >
assertEquals(${pt.boxed}.NaN, percentile(1.0, new ${pt.primitive}[]{4,2,3, ${pt.boxed}.NaN}));

assertEquals(${pt.boxed}.NEGATIVE_INFINITY, percentile(0.0, new ${pt.primitive}[]{4,2,3, ${pt.boxed}.POSITIVE_INFINITY, ${pt.null}, ${pt.null}, ${pt.boxed}.NEGATIVE_INFINITY}));
assertEquals((${pt.primitive})2, percentile(0.25, new ${pt.primitive}[]{4,2,3, ${pt.boxed}.POSITIVE_INFINITY, ${pt.null}, ${pt.null}, ${pt.boxed}.NEGATIVE_INFINITY}));
assertEquals((${pt.primitive})3, percentile(0.5, new ${pt.primitive}[]{4,2,3, ${pt.boxed}.POSITIVE_INFINITY, ${pt.null}, ${pt.null}, ${pt.boxed}.NEGATIVE_INFINITY}));
assertEquals((${pt.primitive})4, percentile(0.75, new ${pt.primitive}[]{4,2,3, ${pt.boxed}.POSITIVE_INFINITY, ${pt.null}, ${pt.null}, ${pt.boxed}.NEGATIVE_INFINITY}));
assertEquals(${pt.boxed}.POSITIVE_INFINITY, percentile(1.0, new ${pt.primitive}[]{4,2,3, ${pt.boxed}.POSITIVE_INFINITY, ${pt.null}, ${pt.null}, ${pt.boxed}.NEGATIVE_INFINITY}));
</#if>
}

public void test${pt.boxed}Wsum() {
Expand Down
11 changes: 7 additions & 4 deletions engine/table/src/main/java/io/deephaven/engine/table/impl/by/CharChunkedVarOperator.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,10 +96,13 @@ private boolean addChunk(CharChunk<? extends Values> values, long destination, i
final double variance = (newSum2 - (newSum * newSum / nonNullCount)) / (nonNullCount - 1);
resultColumn.set(destination, std ? Math.sqrt(variance) : variance);
}
} else if (nonNullCounter.getCountUnsafe(destination) == 1) {
resultColumn.set(destination, Double.NaN);
} else if (nonNullCounter.getCountUnsafe(destination) == 0) {
resultColumn.set(destination, NULL_DOUBLE);
} else {
final long nonNullCount = nonNullCounter.getCountUnsafe(destination);
if (nonNullCount == 0) {
resultColumn.set(destination, NULL_DOUBLE);
} else if (nonNullCount == 1) {
resultColumn.set(destination, Double.NaN);
}
}
return true;
}
Expand Down
23 changes: 13 additions & 10 deletions engine/table/src/main/java/io/deephaven/engine/table/impl/by/FloatChunkedVarOperator.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@
//
package io.deephaven.engine.table.impl.by;

import io.deephaven.base.verify.Assert;
import io.deephaven.chunk.attributes.ChunkLengths;
import io.deephaven.chunk.attributes.ChunkPositions;
import io.deephaven.chunk.attributes.Values;
Expand Down Expand Up @@ -99,7 +100,8 @@ private boolean addChunk(FloatChunk<? extends Values> values, long destination,
sumSource.set(destination, newSum);
sum2Source.set(destination, newSum2);

if (forceNanResult || nonNullCount <= 1) {
Assert.neqZero(nonNullCount, "nonNullCount");
if (forceNanResult || nonNullCount == 1) {
resultColumn.set(destination, Double.NaN);
} else {
// If the sum or sumSquared has reached +/-Infinity, we are stuck with NaN forever.
Expand All @@ -111,15 +113,20 @@ private boolean addChunk(FloatChunk<? extends Values> values, long destination,
resultColumn.set(destination, std ? Math.sqrt(variance) : variance);
}
return true;
} else if (forceNanResult || (nonNullCounter.getCountUnsafe(destination) == 1)) {
} else if (forceNanResult) {
resultColumn.set(destination, Double.NaN);
return true;
} else if (nonNullCounter.getCountUnsafe(destination) == 0) {
resultColumn.set(destination, NULL_DOUBLE);
return true;
} else {
return false;
final long totalNormalCount = nonNullCounter.getCountUnsafe(destination);
if (totalNormalCount == 0) {
resultColumn.set(destination, NULL_DOUBLE);
return true;
} else if (totalNormalCount == 1) {
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
resultColumn.set(destination, Double.NaN);
return true;
}
}
return false;
}

private static double computeVariance(long nonNullCount, double newSum, double newSum2) {
Expand Down Expand Up @@ -175,10 +182,6 @@ private boolean removeChunk(FloatChunk<? extends Values> values, long destinatio
newSum = sumSource.getUnsafe(destination);
newSum2 = sum2Source.getUnsafe(destination);
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}
if (totalNormalCount <= 1) {
resultColumn.set(destination, Double.NaN);
return true;
}

// If the sum has reach +/-Infinity, we are stuck with NaN forever.
if (Double.isInfinite(newSum) || Double.isInfinite(newSum2)) {
Expand Down
Loading