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 2 commits
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
131 changes: 85 additions & 46 deletions engine/function/src/templates/Numeric.ftl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -370,9 +370,11 @@ public class Numeric {
try ( final ${pt.vectorIterator} vi = values.iterator() ) {
while ( vi.hasNext() ) {
final ${pt.primitive} c = vi.${pt.iteratorNext}();
<#if pt.valueType.isFloat >
if (isNaN(c)) {
return Double.NaN;
}
</#if>
if (!isNull(c)) {
sum += c;
count++;
Expand Down Expand Up @@ -431,12 +433,14 @@ public class Numeric {
try ( final ${pt.vectorIterator} vi = values.iterator() ) {
while ( vi.hasNext() ) {
final ${pt.primitive} c = vi.${pt.iteratorNext}();
<#if pt.valueType.isFloat >
if (isNaN(c)) {
return Double.NaN;
}
if (isInf(c)) {
return Double.POSITIVE_INFINITY;
}
</#if>
if (!isNull(c)) {
sum += Math.abs(c);
count++;
Expand Down Expand Up @@ -504,9 +508,11 @@ public class Numeric {
try ( final ${pt.vectorIterator} vi = values.iterator() ) {
while ( vi.hasNext() ) {
final ${pt.primitive} c = vi.${pt.iteratorNext}();
<#if pt.valueType.isFloat >
if (isNaN(c) || isInf(c)) {
return Double.NaN;
}
</#if>
if (!isNull(c)) {
sum += (double)c;
sum2 += (double)c * (double)c;
Expand Down Expand Up @@ -619,6 +625,7 @@ public class Numeric {
double count = 0;
double count2 = 0;
long nullCount = 0;
long valueCount = 0;

try (
final ${pt.vectorIterator} vi = values.iterator();
Expand All @@ -627,17 +634,22 @@ public class Numeric {
while (vi.hasNext()) {
final ${pt.primitive} c = vi.${pt.iteratorNext}();
final ${pt2.primitive} w = wi.${pt2.iteratorNext}();
<#if pt.valueType.isFloat >
if (isNaN(c) || isInf(c)) {
return Double.NaN;
}
</#if>
<#if pt2.valueType.isFloat >
if (isNaN(w) || isInf(w)) {
return Double.NaN;
}
</#if>
if (!isNull(c) && !isNull(w)) {
sum += w * c;
sum2 += w * c * c;
count += w;
count2 += w * w;
valueCount++;
} else {
nullCount++;
}
Expand All @@ -649,7 +661,7 @@ public class Numeric {
}

// Return NaN if overflow or too few values to compute variance.
if (count <= 1 || Double.isInfinite(sum) || Double.isInfinite(sum2)) {
if (valueCount <= 1 || Double.isInfinite(sum) || Double.isInfinite(sum2)) {
return Double.NaN;
}

Expand Down Expand Up @@ -921,7 +933,7 @@ public class Numeric {
}
}

return s == NULL_DOUBLE ? NULL_DOUBLE : s * Math.sqrt(sumw2/sumw/sumw);
return s * Math.sqrt(sumw2/sumw/sumw);
}

</#if>
Expand Down Expand Up @@ -1280,52 +1292,64 @@ public class Numeric {
if (isNaN(sorted[sorted.length - 1])) {
return Double.NaN; // Any NaN will pollute the result and NaN always sorted to the end.
}
</#if>

int nullStart = -1;
int nullCount = 0;

for (int i = 0; i < n; i++) {
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
final ${pt.primitive} val = sorted[i];
if (val > ${pt.null}) {
break; // no more NULL possible
} else if (nullStart == -1) {
if (isNull(val)) {
}
if (isNull(val)) {
nullCount++;
if (nullStart == -1) {
nullStart = i;
nullCount++;
}
} else if (isNull(val)) {
nullCount++;
}
}
<#else>
int nullCount = 0;
for (int i = 0; i < n && isNull(sorted[i]); i++) {
nullCount++;
<#if pt.valueType.isFloat >
if (nullStart == -1) {
nullStart = i;
}
</#if>
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}
</#if>

if (nullCount == n) {
return NULL_DOUBLE;
} else if (nullCount > 0) {
}
if (nullCount > 0) {
n -= nullCount;
final int medianIndex = n / 2;
if (n % 2 == 0) {
<#if pt.valueType.isFloat >
if (nullStart > 0) {
final int medianIndex = n / 2;
if (n % 2 == 0) {
final int idx1 = (medianIndex - 1) >= nullStart ? (medianIndex - 1) + nullCount : (medianIndex - 1);
final int idx2 = medianIndex >= nullStart ? medianIndex + nullCount : medianIndex;
return 0.5 * (sorted[idx1] + sorted[idx2]);
}
final int adjustedIndex = medianIndex >= nullStart ? medianIndex + nullCount : medianIndex;
return sorted[adjustedIndex];
}
final int idx1 = (medianIndex - 1) < nullStart ? (medianIndex - 1) : (medianIndex - 1) + nullCount;
final int idx2 = medianIndex < nullStart ? medianIndex : medianIndex + nullCount;
<#else>
final int idx1 = (medianIndex - 1) + nullCount;
final int idx2 = medianIndex + nullCount;
</#if>
if (n % 2 == 0) {
int index = n / 2;
return 0.5 * (sorted[n / 2 - 1 + nullCount] + sorted[n / 2 + nullCount]);
return 0.5 * (sorted[idx1] + sorted[idx2]);
}
return sorted[n / 2 + nullCount];
<#if pt.valueType.isFloat >
final int adjustedIndex = medianIndex < nullStart ? medianIndex : medianIndex + nullCount;
<#else>
final int adjustedIndex = medianIndex + nullCount;
</#if>
return sorted[adjustedIndex];
}
if (n % 2 == 0)
return 0.5 * (sorted[n / 2 - 1] + sorted[n / 2]);
return sorted[n / 2];
final int medianIndex = n / 2;
if (n % 2 == 0) {
return 0.5 * (sorted[medianIndex - 1] + sorted[medianIndex]);
}
return sorted[medianIndex];
}


lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
/**
* Returns the percentile. {@code null} input values are ignored but {@code NaN} values will poison the computation,
* and {@code NaN} will be returned
Expand Down Expand Up @@ -1361,49 +1385,52 @@ public class Numeric {

int n = values.intSize("percentile");

if (n == 0) {
return ${pt.null};
}

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

<#if pt.valueType.isFloat >
if (isNaN(sorted[sorted.length - 1])) {
return ${pt.boxed}.NaN; // Any NaN will pollute the result and NaN always sorted to the end.
}
</#if>

int nullStart = -1;
int nullCount = 0;

for (int i = 0; i < n; i++) {
final ${pt.primitive} val = sorted[i];
if (val > ${pt.null}) {
break; // no more NULL possible
} else if (nullStart == -1) {
if (isNull(val)) {
}
if (isNull(val)) {
nullCount++;
if (nullStart == -1) {
nullStart = i;
nullCount++;
}
} else if (isNull(val)) {
nullCount++;
}
}
<#else>
int nullCount = 0;
for (int i = 0; i < n && isNull(sorted[i]); i++) {
nullCount++;
<#if pt.valueType.isFloat >
if (nullStart == -1) {
nullStart = i;
}
</#if>
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
}
</#if>

if (nullCount == n) {
return ${pt.null};
} else if (nullCount > 0) {
}
if (nullCount > 0) {
n -= nullCount;
<#if pt.valueType.isFloat >
if (nullStart > 0) {
int idx = (int) Math.round(percentile * (n - 1));
final int adjustedIndex = idx >= nullStart ? idx + nullCount : idx;
return sorted[adjustedIndex];
}
</#if>
final int idx = (int) Math.round(percentile * (n - 1));
final int adjustedIndex = idx >= nullStart ? idx + nullCount : idx;
lbooker42 marked this conversation as resolved.
Show resolved Hide resolved
return sorted[adjustedIndex];
<#else>
int idx = (int) Math.round(percentile * (n - 1));
return sorted[idx + nullCount];
</#if>
}
int idx = (int) Math.round(percentile * (n - 1));
return sorted[idx];
Expand Down Expand Up @@ -1487,12 +1514,16 @@ public class Numeric {
while (v0i.hasNext()) {
final ${pt.primitive} v0 = v0i.${pt.iteratorNext}();
final ${pt2.primitive} v1 = v1i.${pt2.iteratorNext}();
<#if pt.valueType.isFloat >
if (isNaN(v0) || isInf(v0)) {
return Double.NaN;
}
</#if>
<#if pt2.valueType.isFloat >
if (isNaN(v1) || isInf(v1)) {
return Double.NaN;
}
</#if>

if (!isNull(v0) && !isNull(v1)) {
sum0 += v0;
Expand Down Expand Up @@ -1588,12 +1619,16 @@ public class Numeric {
while (v0i.hasNext()) {
final ${pt.primitive} v0 = v0i.${pt.iteratorNext}();
final ${pt2.primitive} v1 = v1i.${pt2.iteratorNext}();
<#if pt.valueType.isFloat >
if (isNaN(v0) || isInf(v0)) {
return Double.NaN;
}
</#if>
<#if pt2.valueType.isFloat >
if (isNaN(v1) || isInf(v1)) {
return Double.NaN;
}
</#if>

if (!isNull(v0) && !isNull(v1)) {
sum0 += v0;
Expand Down Expand Up @@ -2904,12 +2939,16 @@ public class Numeric {
while (vi.hasNext()) {
final ${pt.primitive} c = vi.${pt.iteratorNext}();
final ${pt2.primitive} w = wi.${pt2.iteratorNext}();
<#if pt.valueType.isFloat >
if (isNaN(c)) {
return Double.NaN;
}
</#if>
<#if pt2.valueType.isFloat >
if (isNaN(w)) {
return Double.NaN;
}
</#if>
if (!isNull(c) && !isNull(w)) {
vsum += c * w;
wsum += w;
Expand Down
Loading
Loading