org.datavec.dataframe.api.FloatColumn Maven / Gradle / Ivy
Show all versions of datavec-dataframe Show documentation
package org.datavec.dataframe.api;
import com.google.common.base.Preconditions;
import com.google.common.base.Strings;
import it.unimi.dsi.fastutil.floats.*;
import it.unimi.dsi.fastutil.ints.IntComparator;
import org.datavec.dataframe.columns.AbstractColumn;
import org.datavec.dataframe.columns.Column;
import org.datavec.dataframe.filtering.FloatBiPredicate;
import org.datavec.dataframe.filtering.FloatPredicate;
import org.datavec.dataframe.io.TypeUtils;
import org.datavec.dataframe.reducing.NumericReduceUtils;
import org.datavec.dataframe.store.ColumnMetadata;
import org.datavec.dataframe.util.BitmapBackedSelection;
import org.datavec.dataframe.util.Selection;
import org.datavec.dataframe.util.Stats;
import java.nio.ByteBuffer;
import java.util.Arrays;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
import static org.datavec.dataframe.columns.FloatColumnUtils.*;
import static org.datavec.dataframe.reducing.NumericReduceUtils.*;
/**
* A column in a base table that contains float values
*/
public class FloatColumn extends AbstractColumn implements FloatIterable, NumericColumn {
public static final float MISSING_VALUE = (float) ColumnType.FLOAT.getMissingValue();
private static final int BYTE_SIZE = 4;
private static int DEFAULT_ARRAY_SIZE = 128;
private FloatArrayList data;
public FloatColumn(String name) {
super(name);
data = new FloatArrayList(DEFAULT_ARRAY_SIZE);
}
public FloatColumn(String name, int initialSize) {
super(name);
data = new FloatArrayList(initialSize);
}
public FloatColumn(ColumnMetadata metadata) {
super(metadata);
data = new FloatArrayList(metadata.getSize());
}
public int size() {
return data.size();
}
@Override
public Table summary() {
return stats().asTable();
}
public Stats stats() {
return Stats.create(this);
}
@Override
public int countUnique() {
FloatSet floats = new FloatOpenHashSet();
for (int i = 0; i < size(); i++) {
floats.add(data.getFloat(i));
}
return floats.size();
}
/**
* Returns the largest ("top") n values in the column
*
* @param n The maximum number of records to return. The actual number will be smaller if n is greater than the
* number of observations in the column
* @return A list, possibly empty, of the largest observations
*/
public FloatArrayList top(int n) {
FloatArrayList top = new FloatArrayList();
float[] values = data.toFloatArray();
FloatArrays.parallelQuickSort(values, reverseFloatComparator);
for (int i = 0; i < n && i < values.length; i++) {
top.add(values[i]);
}
return top;
}
/**
* Returns the smallest ("bottom") n values in the column
*
* @param n The maximum number of records to return. The actual number will be smaller if n is greater than the
* number of observations in the column
* @return A list, possibly empty, of the smallest n observations
*/
public FloatArrayList bottom(int n) {
FloatArrayList bottom = new FloatArrayList();
float[] values = data.toFloatArray();
FloatArrays.parallelQuickSort(values);
for (int i = 0; i < n && i < values.length; i++) {
bottom.add(values[i]);
}
return bottom;
}
@Override
public FloatColumn unique() {
FloatSet floats = new FloatOpenHashSet();
for (int i = 0; i < size(); i++) {
floats.add(data.getFloat(i));
}
FloatColumn column = new FloatColumn(name() + " Unique values", floats.size());
floats.forEach(column::add);
return column;
}
public FloatArrayList data() {
return data;
}
@Override
public ColumnType type() {
return ColumnType.FLOAT;
}
public float firstElement() {
if (size() > 0) {
return data.getFloat(0);
}
return MISSING_VALUE;
}
// Reduce functions applied to the whole column
public double sum() {
return sum.reduce(this);
}
public double product() {
return product.reduce(this);
}
public double mean() {
return mean.reduce(this);
}
public double median() {
return median.reduce(this);
}
public double quartile1() {
return quartile1.reduce(this);
}
public double quartile3() {
return quartile3.reduce(this);
}
public double percentile(double percentile) {
return NumericReduceUtils.percentile(this.toDoubleArray(), percentile);
}
public double range() {
return range.reduce(this);
}
public double max() {
return max.reduce(this);
}
public double min() {
return min.reduce(this);
}
public double variance() {
return variance.reduce(this);
}
public double populationVariance() {
return populationVariance.reduce(this);
}
public double standardDeviation() {
return stdDev.reduce(this);
}
public double sumOfLogs() {
return sumOfLogs.reduce(this);
}
public double sumOfSquares() {
return sumOfSquares.reduce(this);
}
public double geometricMean() {
return geometricMean.reduce(this);
}
/**
* Returns the quadraticMean, aka the root-mean-square, for all values in this column
*/
public double quadraticMean() {
return quadraticMean.reduce(this);
}
public double kurtosis() {
return kurtosis.reduce(this);
}
public double skewness() {
return skewness.reduce(this);
}
/**
* Adds the given float to this column
*/
public void add(float f) {
data.add(f);
}
/**
* Adds the given double to this column, after casting it to a float
*/
public void add(double d) {
data.add((float) d);
}
// Predicate functions
public Selection isLessThan(float f) {
return select(isLessThan, f);
}
public Selection isMissing() {
return select(isMissing);
}
public Selection isNotMissing() {
return select(isNotMissing);
}
public Selection isGreaterThan(float f) {
return select(isGreaterThan, f);
}
public Selection isGreaterThanOrEqualTo(float f) {
return select(isGreaterThanOrEqualTo, f);
}
public Selection isLessThanOrEqualTo(float f) {
return select(isLessThanOrEqualTo, f);
}
public Selection isNotEqualTo(float f) {
return select(notIsEqualTo, f);
}
public Selection isEqualTo(float f) {
return select(isEqualTo, f);
}
public Selection isEqualTo(FloatColumn f) {
Selection results = new BitmapBackedSelection();
int i = 0;
FloatIterator floatIterator = f.iterator();
for (float floats : data) {
if (floats == floatIterator.nextFloat()) {
results.add(i);
}
i++;
}
return results;
}
@Override
public String getString(int row) {
return String.valueOf(data.getFloat(row));
}
@Override
public FloatColumn emptyCopy() {
FloatColumn column = new FloatColumn(name());
column.setComment(comment());
return column;
}
@Override
public FloatColumn emptyCopy(int rowSize) {
FloatColumn column = new FloatColumn(name(), rowSize);
column.setComment(comment());
return column;
}
@Override
public void clear() {
data = new FloatArrayList(DEFAULT_ARRAY_SIZE);
}
@Override
public FloatColumn copy() {
FloatColumn column = FloatColumn.create(name(), data);
column.setComment(comment());
return column;
}
@Override
public void sortAscending() {
Arrays.parallelSort(data.elements());
}
@Override
public void sortDescending() {
FloatArrays.parallelQuickSort(data.elements(), reverseFloatComparator);
}
@Override
public boolean isEmpty() {
return data.isEmpty();
}
public static FloatColumn create(String name) {
return new FloatColumn(name);
}
public static FloatColumn create(String name, int initialSize) {
return new FloatColumn(name, initialSize);
}
public static FloatColumn create(String name, FloatArrayList floats) {
FloatColumn column = new FloatColumn(name, floats.size());
column.data = new FloatArrayList(floats.size());
column.data.addAll(floats);
return column;
}
/**
* Compares two floats, such that a sort based on this comparator would sort in descending order
*/
FloatComparator reverseFloatComparator = new FloatComparator() {
@Override
public int compare(Float o2, Float o1) {
return (o1 < o2 ? -1 : (o1.equals(o2) ? 0 : 1));
}
@Override
public int compare(float o2, float o1) {
return (o1 < o2 ? -1 : (o1 == o2 ? 0 : 1));
}
};
/**
* Returns the count of missing values in this column
*
* Implementation note: We use NaN for missing, so we can't compare against the MISSING_VALUE and use val != val instead
*/
@Override
public int countMissing() {
int count = 0;
for (int i = 0; i < size(); i++) {
float f = get(i);
if (f != f) {
count++;
}
}
return count;
}
@Override
public void addCell(String object) {
try {
add(convert(object));
} catch (NumberFormatException nfe) {
throw new NumberFormatException(name() + ": " + nfe.getMessage());
} catch (NullPointerException e) {
throw new RuntimeException(name() + ": " + String.valueOf(object) + ": " + e.getMessage());
}
}
/**
* Returns a float that is parsed from the given String
*
* We remove any commas before parsing
*/
public static float convert(String stringValue) {
if (Strings.isNullOrEmpty(stringValue) || TypeUtils.MISSING_INDICATORS.contains(stringValue)) {
return MISSING_VALUE;
}
Matcher matcher = COMMA_PATTERN.matcher(stringValue);
return Float.parseFloat(matcher.replaceAll(""));
}
/**
* Returns the natural log of the values in this column as a new FloatColumn
*/
public FloatColumn logN() {
FloatColumn newColumn = FloatColumn.create(name() + "[logN]", size());
for (float value : this) {
newColumn.add((float) Math.log(value));
}
return newColumn;
}
public FloatColumn log10() {
FloatColumn newColumn = FloatColumn.create(name() + "[log10]", size());
for (float value : this) {
newColumn.add((float) Math.log10(value));
}
return newColumn;
}
/**
* Returns the natural log of the values in this column, after adding 1 to each so that zero
* values don't return -Infinity
*/
public FloatColumn log1p() {
FloatColumn newColumn = FloatColumn.create(name() + "[1og1p]", size());
for (float value : this) {
newColumn.add((float) Math.log1p(value));
}
return newColumn;
}
public FloatColumn round() {
FloatColumn newColumn = FloatColumn.create(name() + "[rounded]", size());
for (float value : this) {
newColumn.add(Math.round(value));
}
return newColumn;
}
public FloatColumn abs() {
FloatColumn newColumn = FloatColumn.create(name() + "[abs]", size());
for (float value : this) {
newColumn.add(Math.abs(value));
}
return newColumn;
}
public FloatColumn square() {
FloatColumn newColumn = FloatColumn.create(name() + "[sq]", size());
for (float value : this) {
newColumn.add(value * value);
}
return newColumn;
}
public FloatColumn sqrt() {
FloatColumn newColumn = FloatColumn.create(name() + "[sqrt]", size());
for (float value : this) {
newColumn.add((float) Math.sqrt(value));
}
return newColumn;
}
public FloatColumn cubeRoot() {
FloatColumn newColumn = FloatColumn.create(name() + "[cbrt]", size());
for (float value : this) {
newColumn.add((float) Math.cbrt(value));
}
return newColumn;
}
public FloatColumn cube() {
FloatColumn newColumn = FloatColumn.create(name() + "[cb]", size());
for (float value : this) {
newColumn.add(value * value * value);
}
return newColumn;
}
public FloatColumn remainder(FloatColumn column2) {
FloatColumn result = FloatColumn.create(name() + " % " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) % column2.get(r));
}
return result;
}
public FloatColumn add(FloatColumn column2) {
FloatColumn result = FloatColumn.create(name() + " + " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) + column2.get(r));
}
return result;
}
public FloatColumn subtract(FloatColumn column2) {
FloatColumn result = FloatColumn.create(name() + " - " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) - column2.get(r));
}
return result;
}
public FloatColumn multiply(FloatColumn column2) {
FloatColumn result = FloatColumn.create(name() + " * " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) * column2.get(r));
}
return result;
}
public FloatColumn multiply(IntColumn column2) {
FloatColumn result = FloatColumn.create(name() + " * " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) * column2.get(r));
}
return result;
}
public FloatColumn multiply(LongColumn column2) {
FloatColumn result = FloatColumn.create(name() + " * " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) * column2.get(r));
}
return result;
}
public FloatColumn multiply(ShortColumn column2) {
FloatColumn result = FloatColumn.create(name() + " * " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) * column2.get(r));
}
return result;
}
public FloatColumn divide(FloatColumn column2) {
FloatColumn result = FloatColumn.create(name() + " / " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) / column2.get(r));
}
return result;
}
public FloatColumn divide(IntColumn column2) {
FloatColumn result = FloatColumn.create(name() + " / " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) / column2.get(r));
}
return result;
}
public FloatColumn divide(LongColumn column2) {
FloatColumn result = FloatColumn.create(name() + " / " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) / column2.get(r));
}
return result;
}
public FloatColumn divide(ShortColumn column2) {
FloatColumn result = FloatColumn.create(name() + " / " + column2.name(), size());
for (int r = 0; r < size(); r++) {
result.add(get(r) / column2.get(r));
}
return result;
}
/**
* For each item in the column, returns the same number with the sign changed.
* For example:
* -1.3 returns 1.3,
* 2.135 returns -2.135
* 0 returns 0
*/
public FloatColumn neg() {
FloatColumn newColumn = FloatColumn.create(name() + "[neg]", size());
for (float value : this) {
newColumn.add(value * -1);
}
return newColumn;
}
private static final Pattern COMMA_PATTERN = Pattern.compile(",");
/**
* Compares the given ints, which refer to the indexes of the floats in this column, according to the values of the
* floats themselves
*/
@Override
public IntComparator rowComparator() {
return comparator;
}
private final IntComparator comparator = new IntComparator() {
@Override
public int compare(Integer r1, Integer r2) {
float f1 = data.getFloat(r1);
float f2 = data.getFloat(r2);
return Float.compare(f1, f2);
}
public int compare(int r1, int r2) {
float f1 = data.getFloat(r1);
float f2 = data.getFloat(r2);
return Float.compare(f1, f2);
}
};
public float get(int index) {
return data.getFloat(index);
}
@Override
public float getFloat(int index) {
return data.getFloat(index);
}
public void set(int r, float value) {
data.set(r, value);
}
// TODO(lwhite): Reconsider the implementation of this functionality to allow user to provide a specific max error.
// TODO(lwhite): continued: Also see section in Effective Java on floating point comparisons.
Selection isCloseTo(float target) {
Selection results = new BitmapBackedSelection();
int i = 0;
for (float f : data) {
if (Float.compare(f, target) == 0) {
results.add(i);
}
i++;
}
return results;
}
Selection isCloseTo(double target) {
Selection results = new BitmapBackedSelection();
int i = 0;
for (float f : data) {
if (Double.compare(f, 0.0) == 0) {
results.add(i);
}
i++;
}
return results;
}
Selection isPositive() {
return select(isPositive);
}
Selection isNegative() {
return select(isNegative);
}
Selection isNonNegative() {
return select(isNonNegative);
}
public double[] toDoubleArray() {
double[] output = new double[data.size()];
for (int i = 0; i < data.size(); i++) {
output[i] = data.getFloat(i);
}
return output;
}
public String print() {
StringBuilder builder = new StringBuilder();
builder.append(title());
for (Float aData : data) {
builder.append(String.valueOf(aData));
builder.append('\n');
}
return builder.toString();
}
@Override
public String toString() {
return "Float column: " + name();
}
@Override
public void append(Column column) {
Preconditions.checkArgument(column.type() == this.type());
FloatColumn floatColumn = (FloatColumn) column;
for (int i = 0; i < floatColumn.size(); i++) {
add(floatColumn.get(i));
}
}
@Override
public FloatIterator iterator() {
return data.iterator();
}
public Selection select(FloatPredicate predicate) {
Selection bitmap = new BitmapBackedSelection();
for (int idx = 0; idx < data.size(); idx++) {
float next = data.getFloat(idx);
if (predicate.test(next)) {
bitmap.add(idx);
}
}
return bitmap;
}
public Selection select(FloatBiPredicate predicate, float value) {
Selection bitmap = new BitmapBackedSelection();
for (int idx = 0; idx < data.size(); idx++) {
float next = data.getFloat(idx);
if (predicate.test(next, value)) {
bitmap.add(idx);
}
}
return bitmap;
}
FloatSet asSet() {
return new FloatOpenHashSet(data);
}
public boolean contains(float value) {
return data.contains(value);
}
@Override
public int byteSize() {
return BYTE_SIZE;
}
/**
* Returns the contents of the cell at rowNumber as a byte[]
*/
@Override
public byte[] asBytes(int rowNumber) {
return ByteBuffer.allocate(4).putFloat(get(rowNumber)).array();
}
@Override
public FloatColumn difference() {
FloatColumn returnValue = new FloatColumn(this.name(), this.size());
returnValue.add(FloatColumn.MISSING_VALUE);
for (int current = 0; current < this.size(); current++) {
if (current + 1 < this.size()) {
/*
* check for missing values:
* note that for floats you test val != val,
* since a missing float is encoded as Float.NaN and nothing is equal to NaN.
*/
float currentValue = get(current);
float nextValue = get(current + 1);
if (currentValue != currentValue || nextValue != nextValue) {
returnValue.add(Float.NaN);
} else {
returnValue.add(nextValue - currentValue);
}
}
}
return returnValue;
}
}