org.datavec.dataframe.io.csv.CsvReader Maven / Gradle / Ivy
Show all versions of datavec-dataframe Show documentation
package org.datavec.dataframe.io.csv;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Strings;
import com.google.common.collect.Lists;
import com.opencsv.CSVReader;
import edu.umd.cs.findbugs.annotations.Nullable;
import net.jcip.annotations.Immutable;
import org.apache.commons.lang3.StringUtils;
import org.datavec.dataframe.api.ColumnType;
import org.datavec.dataframe.api.Table;
import org.datavec.dataframe.columns.Column;
import org.datavec.dataframe.io.TypeUtils;
import java.io.*;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.time.LocalDate;
import java.time.LocalDateTime;
import java.time.LocalTime;
import java.time.format.DateTimeParseException;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.CopyOnWriteArrayList;
import java.util.function.Predicate;
import static org.datavec.dataframe.api.ColumnType.*;
/**
*
*/
@Immutable
public class CsvReader {
/**
* Constructs and returns a table from one or more CSV files, all containing the same column types
*
* This constructor assumes the files have a one-line header, which is used to populate the column names,
* and that they use a comma to separate between columns.
*
* @throws IOException If there is an issue reading any of the files
*/
public static Table read(ColumnType types[], String... fileNames) throws IOException {
if (fileNames.length == 1) {
return read(types, true, ',', fileNames[0]);
} else {
Table table = read(types, true, ',', fileNames[0]);
for (int i = 1; i < fileNames.length; i++) {
String fileName = fileNames[i];
table.append(read(types, true, ',', fileName));
}
return table;
}
}
/**
* Returns a Table constructed from a CSV File with the given file name
*
* The @code{fileName} is used as the initial table name for the new table
*
* @param types An array of the types of columns in the file, in the order they appear
* @param header Is the first row in the file a header?
* @param columnSeparator the delimiter
* @param fileName The fully specified file name. It is used to provide a default name for the table
* @return A Table containing the data in the csv file.
* @throws IOException
*/
public static Table read(ColumnType types[], boolean header, char columnSeparator, String fileName)
throws IOException {
InputStream stream = new FileInputStream(fileName);
return read(fileName, types, header, columnSeparator, stream);
}
/**
* Retuns the given file after autodetecting the column types, or trying to
*
* @param fileName The name of the file to load
* @param header True if the file has a single header row. False if it has no header row.
* Multi-line headers are not supported
* @param delimiter a char that divides the columns in the source file, often a comma or tab
* @return A table containing the data from the file
* @throws IOException
*/
public static Table read(String fileName, boolean header, char delimiter) throws IOException {
ColumnType[] columnTypes = detectColumnTypes(fileName, header, delimiter);
return read(columnTypes, true, delimiter, fileName);
}
public static Table read(String tableName, ColumnType types[], boolean header, char columnSeparator,
InputStream stream) throws IOException {
BufferedReader streamReader = new BufferedReader(new InputStreamReader(stream));
Table table;
try (CSVReader reader = new CSVReader(streamReader, columnSeparator, '"')) {
String[] nextLine;
String[] columnNames;
List headerRow;
if (header) {
nextLine = reader.readNext();
headerRow = Lists.newArrayList(nextLine);
columnNames = selectColumnNames(headerRow, types);
} else {
columnNames = makeColumnNames(types);
headerRow = Lists.newArrayList(columnNames);
}
table = Table.create(nameMaker(tableName));
for (int x = 0; x < types.length; x++) {
if (types[x] != ColumnType.SKIP) {
String columnName = headerRow.get(x);
if (Strings.isNullOrEmpty(columnName)) {
columnName = "Column " + table.columnCount();
}
Column newColumn = TypeUtils.newColumn(columnName.trim(), types[x]);
table.addColumn(newColumn);
}
}
int[] columnIndexes = new int[columnNames.length];
for (int i = 0; i < columnIndexes.length; i++) {
// get the index in the original table, which includes skipped fields
columnIndexes[i] = headerRow.indexOf(columnNames[i]);
}
// Add the rows
long rowNumber = header ? 1L : 0L;
while ((nextLine = reader.readNext()) != null) {
// for each column that we're including (not skipping)
int cellIndex = 0;
for (int columnIndex : columnIndexes) {
Column column = table.column(cellIndex);
try {
column.addCell(nextLine[columnIndex]);
} catch (Exception e) {
throw new AddCellToColumnException(e, columnIndex, rowNumber, columnNames, nextLine);
}
cellIndex++;
}
rowNumber++;
}
}
return table;
}
/**
* Returns a Table constructed from a CSV File with the given file name
*
* The @code{fileName} is used as the initial table name for the new table
*
* @param types An array of the types of columns in the file, in the order they appear
* @param header Is the first row in the file a header?
* @param columnSeparator the delimiter
* @param name The fully specified file name. It is used to provide a default name for the table
* @return A Table containing the data in the csv file.
* @throws IOException
*/
public static Table headerOnly(String name, ColumnType types[], boolean header, char columnSeparator,
InputStream stream) throws IOException {
BufferedReader streamReader = new BufferedReader(new InputStreamReader(stream));
Table table;
try (CSVReader reader = new CSVReader(streamReader, columnSeparator, '"')) {
String[] nextLine;
String[] columnNames;
List headerRow;
if (header) {
nextLine = reader.readNext();
headerRow = Lists.newArrayList(nextLine);
columnNames = selectColumnNames(headerRow, types);
} else {
columnNames = makeColumnNames(types);
headerRow = Lists.newArrayList(columnNames);
}
table = Table.create(nameMaker(name));
for (int x = 0; x < types.length; x++) {
if (types[x] != ColumnType.SKIP) {
Column newColumn = TypeUtils.newColumn(headerRow.get(x).trim(), types[x]);
table.addColumn(newColumn);
}
}
int[] columnIndexes = new int[columnNames.length];
for (int i = 0; i < columnIndexes.length; i++) {
// get the index in the original table, which includes skipped fields
columnIndexes[i] = headerRow.indexOf(columnNames[i]);
}
// Add the rows
while ((nextLine = reader.readNext()) != null) {
// for each column that we're including (not skipping)
int cellIndex = 0;
for (int columnIndex : columnIndexes) {
Column column = table.column(cellIndex);
column.addCell(nextLine[columnIndex]);
cellIndex++;
}
}
}
return table;
}
/**
* Returns a Table constructed from a CSV File with the given file name
*
* The @code{fileName} is used as the initial table name for the new table
*
* @param types An array of the types of columns in the file, in the order they appear
* @param header Is the first row in the file a header?
* @param columnSeparator the delimiter
* @param fileName The fully specified file name. It is used to provide a default name for the table
* @return A Relation containing the data in the csv file.
* @throws IOException
*/
public static Table headerOnly(ColumnType types[], boolean header, char columnSeparator, String fileName)
throws IOException {
InputStream stream = new FileInputStream(fileName);
return headerOnly(fileName, types, header, columnSeparator, stream);
}
/**
* Returns the structure of the table given by {@code csvFileName} as detected by analysis of a sample of the data
*
* @throws IOException
*/
public static Table detectedColumnTypes(String csvFileName, boolean header, char delimiter) throws IOException {
ColumnType[] types = detectColumnTypes(csvFileName, header, delimiter);
Table t = headerOnly(types, header, delimiter, csvFileName);
return t.structure();
}
/**
* Returns a string representation of the file types in file {@code csvFilename},
* as determined by the type-detection algorithm
*
* This method is intended to help analysts quickly fix any erroneous types, by printing out the types in a format
* such that they can be edited to correct any mistakes, and used in an array literal
*
* For example:
*
* LOCAL_DATE, // 0 date
* SHORT_INT, // 1 approval
* CATEGORY, // 2 who
*
* Note that the types are array separated, and that the index position and the column name are printed such that
* they
* would be interpreted as comments if you paste the output into an array:
*
* ColumnType[] types = {
* LOCAL_DATE, // 0 date
* SHORT_INT, // 1 approval
* CATEGORY, // 2 who
* }
*
* @throws IOException
*/
public static String printColumnTypes(String csvFileName, boolean header, char delimiter) throws IOException {
Table structure = detectedColumnTypes(csvFileName, header, delimiter);
StringBuilder buf = new StringBuilder();
buf.append("ColumnType[] columnTypes = {");
buf.append('\n');
Column typeCol = structure.column("Column Type");
Column indxCol = structure.column("Index");
Column nameCol = structure.column("Column Name");
// add the column headers
int typeColIndex = structure.columnIndex(typeCol);
int indxColIndex = structure.columnIndex(indxCol);
int nameColIndex = structure.columnIndex(nameCol);
int typeColWidth = typeCol.columnWidth();
int indxColWidth = indxCol.columnWidth();
int nameColWidth = nameCol.columnWidth();
for (int r = 0; r < structure.rowCount(); r++) {
String cell = StringUtils.rightPad(structure.get(typeColIndex, r) + ",", typeColWidth);
buf.append(cell);
buf.append(" // ");
cell = StringUtils.rightPad(structure.get(indxColIndex, r), indxColWidth);
buf.append(cell);
buf.append(' ');
cell = StringUtils.rightPad(structure.get(nameColIndex, r), nameColWidth);
buf.append(cell);
buf.append(' ');
buf.append('\n');
}
buf.append("}");
buf.append('\n');
return buf.toString();
}
/**
* Reads column names from header, skipping any for which the type == SKIP
*/
private static String[] selectColumnNames(List names, ColumnType types[]) {
List header = new ArrayList<>();
for (int i = 0; i < types.length; i++) {
if (types[i] != ColumnType.SKIP) {
header.add(names.get(i));
}
}
String[] result = new String[header.size()];
return header.toArray(result);
}
private static String nameMaker(String path) {
Path p = Paths.get(path);
return p.getFileName().toString();
}
/**
* Provides placeholder column names for when the file read has no header
*/
private static String[] makeColumnNames(ColumnType types[]) {
String[] header = new String[types.length];
for (int i = 0; i < types.length; i++) {
header[i] = "C" + i;
}
return header;
}
@VisibleForTesting
/**
* Estimates and returns the type for each column in the delimited text file {@code file}
*
* The type is determined by checking a sample of the data in the file. Because only a sample of the data is checked,
* the types may be incorrect. If that is the case a Parse Exception will be thrown.
*
* The method {@code printColumnTypes()} can be used to print a list of the detected columns that can be corrected and
* used to explicitely specify the correct column types.
*/
static ColumnType[] detectColumnTypes(String file, boolean header, char delimiter) throws IOException {
int linesToSkip = header ? 1 : 0;
// to hold the results
List columnTypes = new ArrayList<>();
// to hold the data read from the file
List> columnData = new ArrayList<>();
int rowCount = 0; // make sure we don't go over maxRows
try (CSVReader reader = new CSVReader(new FileReader(file), delimiter, '"', linesToSkip)) {
String[] nextLine;
int nextRow = 0;
while ((nextLine = reader.readNext()) != null) {
// initialize the arrays to hold the strings. we don't know how many we need until we read the first row
if (rowCount == 0) {
for (String aNextLine : nextLine) {
columnData.add(new ArrayList<>());
}
}
int columnNumber = 0;
if (rowCount == nextRow) {
for (String field : nextLine) {
columnData.get(columnNumber).add(field);
columnNumber++;
}
}
if (rowCount == nextRow) {
nextRow = nextRow(nextRow);
}
rowCount++;
}
}
// now detect
for (List valuesList : columnData) {
ColumnType detectedType = detectType(valuesList);
columnTypes.add(detectedType);
}
return columnTypes.toArray(new ColumnType[columnTypes.size()]);
}
private static int nextRow(int nextRow) {
if (nextRow < 100) {
return nextRow + 1;
}
if (nextRow < 1000) {
return nextRow + 10;
}
if (nextRow < 10_000) {
return nextRow + 100;
}
if (nextRow < 100_000) {
return nextRow + 1000;
}
if (nextRow < 1_000_000) {
return nextRow + 10_000;
}
if (nextRow < 10_000_000) {
return nextRow + 100_000;
}
if (nextRow < 100_000_000) {
return nextRow + 1_000_000;
}
return nextRow + 10_000_000;
}
private static ColumnType detectType(List valuesList) {
// Types to choose from. When more than one would work, we pick the first of the options
ColumnType[] typeArray = // we leave out category, as that is the default type
{LOCAL_DATE_TIME, LOCAL_TIME, LOCAL_DATE, BOOLEAN, SHORT_INT, INTEGER, LONG_INT, FLOAT};
CopyOnWriteArrayList typeCandidates = new CopyOnWriteArrayList<>(typeArray);
for (String s : valuesList) {
if (Strings.isNullOrEmpty(s) || TypeUtils.MISSING_INDICATORS.contains(s)) {
continue;
}
if (typeCandidates.contains(LOCAL_DATE_TIME)) {
if (!isLocalDateTime.test(s)) {
typeCandidates.remove(LOCAL_DATE_TIME);
}
}
if (typeCandidates.contains(LOCAL_TIME)) {
if (!isLocalTime.test(s)) {
typeCandidates.remove(LOCAL_TIME);
}
}
if (typeCandidates.contains(LOCAL_DATE)) {
if (!isLocalDate.test(s)) {
typeCandidates.remove(LOCAL_DATE);
}
}
if (typeCandidates.contains(BOOLEAN)) {
if (!isBoolean.test(s)) {
typeCandidates.remove(BOOLEAN);
}
}
if (typeCandidates.contains(SHORT_INT)) {
if (!isShort.test(s)) {
typeCandidates.remove(SHORT_INT);
}
}
if (typeCandidates.contains(INTEGER)) {
if (!isInteger.test(s)) {
typeCandidates.remove(INTEGER);
}
}
if (typeCandidates.contains(LONG_INT)) {
if (!isLong.test(s)) {
typeCandidates.remove(LONG_INT);
}
}
if (typeCandidates.contains(FLOAT)) {
if (!isFloat.test(s)) {
typeCandidates.remove(FLOAT);
}
}
}
return selectType(typeCandidates);
}
/**
* Returns the selected candidate for a column of data, by picking the first value in the given list
*
* @param typeCandidates a possibly empty list of candidates. This list should be sorted in order of preference
*/
private static ColumnType selectType(List typeCandidates) {
if (typeCandidates.isEmpty()) {
return CATEGORY;
} else {
return typeCandidates.get(0);
}
}
private static java.util.function.Predicate isBoolean =
s -> TypeUtils.TRUE_STRINGS_FOR_DETECTION.contains(s)
|| TypeUtils.FALSE_STRINGS_FOR_DETECTION.contains(s);
private static Predicate isLong = new Predicate() {
@Override
public boolean test(@Nullable String s) {
try {
Long.parseLong(s);
return true;
} catch (NumberFormatException e) {
// it's all part of the plan
return false;
}
}
};
private static Predicate isInteger = s -> {
try {
Integer.parseInt(s);
return true;
} catch (NumberFormatException e) {
// it's all part of the plan
return false;
}
};
private static Predicate isFloat = s -> {
try {
Float.parseFloat(s);
return true;
} catch (NumberFormatException e) {
// it's all part of the plan
return false;
}
};
private static Predicate isShort = s -> {
try {
Short.parseShort(s);
return true;
} catch (NumberFormatException e) {
// it's all part of the plan
return false;
}
};
private static Predicate isLocalDate = s -> {
try {
LocalDate.parse(s, TypeUtils.DATE_FORMATTER);
return true;
} catch (DateTimeParseException e) {
// it's all part of the plan
return false;
}
};
private static Predicate isLocalTime = s -> {
try {
LocalTime.parse(s, TypeUtils.TIME_DETECTION_FORMATTER);
return true;
} catch (DateTimeParseException e) {
// it's all part of the plan
return false;
}
};
private static Predicate isLocalDateTime = s -> {
try {
LocalDateTime.parse(s, TypeUtils.DATE_TIME_FORMATTER);
return true;
} catch (DateTimeParseException e) {
// it's all part of the plan
return false;
}
};
/**
* Private constructor to prevent instantiation
*/
private CsvReader() {}
}