我的问题实际上至少是重复的：
1) Stop Excel from automatically converting certain text values to dates
2) Excel: Default to TEXT rather than GENERAL when opening a .csv file
Excel 的可能解决方案是 1) 将带有特殊双引号的字段（如 "May 16, 2011"）写入 "=""May 16, 2011""" 或 2) 使用外部数据向导导入 csv/tsv 文件，然后手动选择您想要的列读取为 TEXT 而不是 GENERAL（可以将字段转换为日期）
至于我的用例，我只是使用 Excel 来删除一些列。没有一个解决方案对我有吸引力，因为我不想用特殊引号重写 tsv 文件，而且我有数百列，我不想手动选择每一列以读取为 TEXT。
因此我编写了一个 scala 脚本来按列名过滤 tsv 文件：
package com.jmcejuela.ml
import java.io.InputStream
import java.io.Writer
import scala.io.Codec
import scala.io.Source
import Table._
/**
* Class to represent tables with a fixed size of columns. All rows have the same columns.
*/
class Table(val rows: Seq[Row]) {
lazy val numDiffColumns = rows.foldLeft(Set[Int]())((set, row) => set + row.size)
def toTSV(out: Writer) {
if (rows.isEmpty) out.write(TableEmpty.toString)
else {
out.write(writeLineTSV(rows.head.map(_.name))) //header
rows.foreach(r => out.write(writeLineTSV(r.map(_.value))))
out.close
}
}
/**
* Get a Table with only the given columns.
*/
def filterColumnsByName(columnNames: Set[String]): Table = {
val existingNames = rows.head.map(_.name).toSet
assert(columnNames.forall(n => existingNames.contains(n)), "You want to include column names that do not exist")
new Table(rows.map { row => row.filter(col => columnNames.contains(col.name)) })
}
}
object TableEmpty extends Table(Seq.empty) {
override def toString = "Table(Empty)"
}
object Table {
def apply(rows: Row*) = new Table(rows)
type Row = Array[Column]
/**
* Column representation. Note that each column has a name and a value. Since the class Table
* is a sequence of rows which are a size-fixed array of columns, the name field is redundant
* for Table. However, this column representation could be used in the future to support
* schemata-less tables.
*/
case class Column(name: String, value: String)
private def parseLineTSV(line: String) = line.split("\t")
private def writeLineTSV(line: Seq[String]) = line.mkString("", "\t", "\n")
/**
* It is assumed that the first row gives the names to the columns
*/
def fromTSV(in: InputStream)(implicit encoding: Codec = Codec.UTF8): Table = {
val linesIt = Source.fromInputStream(in).getLines
if (linesIt.isEmpty) TableEmpty
else {
val columnNames = parseLineTSV(linesIt.next)
val padding = {
//add padding of empty columns-fields to lines that do not include last fields because they are empty
def infinite[A](x: A): Stream[A] = x #:: infinite(x)
infinite("")
}
val rows = linesIt.map { line =>
((0 until columnNames.size).zip(parseLineTSV(line) ++: padding).map { case (index, field) => Column(columnNames(index), field) }).toArray
}.toStream
new Table(rows)
}
}
}