All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.datavec.dataframe.io.csv.CsvReader Maven / Gradle / Ivy

Go to download

High-performance Java Dataframe with integrated columnar storage (fork of tablesaw)

There is a newer version: 0.9.1
Show newest version
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() {} }





© 2015 - 2024 Weber Informatics LLC | Privacy Policy