如何压缩 2 个 Deedle 帧并处理缺失值？

Question

给定价格框架 priceFrame 为

                       28881              29021              29399
2010-01-01 00:00:00 -> 123.535878499576   195.28635425580265 189.92210186152082
2010-01-04 00:00:00 -> 124.19087548338847 198.10448102247753 190.1571733631235
2010-01-05 00:00:00 -> 123.82028508465247 197.8259452373992  190.31388769752525
2010-01-06 00:00:00 -> 124.17363872065654 197.80956077945342 189.98478759528152
2010-01-07 00:00:00 -> 123.4583130672824  197.58017836821244 190.31388769752527
2010-01-08 00:00:00 -> 124.23396739021821 198.10448102247756 190.25120196376457
2010-01-11 00:00:00 -> 125.12166067091142 197.87509861123658 190.73701640041008
2010-01-12 00:00:00 -> 124.9234378994945  195.0569718445617  191.41088803833776
2010-01-13 00:00:00 -> 125.06133200134975 195.64681233060992 191.50491663897884
2010-01-14 00:00:00 -> 124.97514818769021 196.28580619049552 191.56760237273951
2010-01-15 00:00:00 -> 123.71686450826103 192.5829186947483  192.08475967626538
2010-01-18 00:00:00 -> 123.71686450826103 194.10667328370621 192.31983117786805
2010-01-19 00:00:00 -> 123.15666971947407 195.87619474185092 191.94371677530378
2010-01-20 00:00:00 -> 121.5622691667727  191.79646471335064 192.82131704795376
2010-01-21 00:00:00 -> 121.5450324040408  188.38849746062752 192.9937028157957
2010-01-22 00:00:00 -> 121.81220222638535 186.8647428716696  192.9937028157957
2010-01-25 00:00:00 -> 121.94147794687466 184.83307008639233 192.9937028157957
2010-01-26 00:00:00 -> 121.38990153945363 185.9799821425972  193.19743145051802
2010-01-27 00:00:00 -> 120.94174570842405 184.91499237612123 193.3541457849198
2010-01-28 00:00:00 -> 120.44187958919875 182.5392459739825  193.22877431739838
2010-01-29 00:00:00 -> 119.4938576389439  183.75169586197052 193.35414578491978

和股息框架 divFrame 为

                       28881     29021     29399
2010-01-04 00:00:00 -> 1.3       <missing> <missing>
2010-01-13 00:00:00 -> <missing> 1.3       <missing>
2010-01-22 00:00:00 -> <missing> <missing> 1.3

我想将它们组合起来，以便在存在股息的情况下获得价格 + 股息，否则保持价格不变。

以下两种尝试

let dfZipped1 = priceFrame.Zip(divFrame, JoinKind.Left, JoinKind.Left, Lookup.Exact, false, fun (p:float) d -> p + d)
        dfZipped1.Print()
        let dfZipped2 = priceFrame.Zip(divFrame, JoinKind.Left, JoinKind.Left, Lookup.Exact, true, fun (p:float) d -> p + d)
        dfZipped2.Print()

结果相同

                       28881              29021              29399
2010-01-01 00:00:00 -> <missing>          <missing>          <missing>
2010-01-04 00:00:00 -> 125.49087548338846 <missing>          <missing>
2010-01-05 00:00:00 -> <missing>          <missing>          <missing>
2010-01-06 00:00:00 -> <missing>          <missing>          <missing>
2010-01-07 00:00:00 -> <missing>          <missing>          <missing>
2010-01-08 00:00:00 -> <missing>          <missing>          <missing>
2010-01-11 00:00:00 -> <missing>          <missing>          <missing>
2010-01-12 00:00:00 -> <missing>          <missing>          <missing>
2010-01-13 00:00:00 -> <missing>          196.94681233060993 <missing>
2010-01-14 00:00:00 -> <missing>          <missing>          <missing>
2010-01-15 00:00:00 -> <missing>          <missing>          <missing>
2010-01-18 00:00:00 -> <missing>          <missing>          <missing>
2010-01-19 00:00:00 -> <missing>          <missing>          <missing>
2010-01-20 00:00:00 -> <missing>          <missing>          <missing>
2010-01-21 00:00:00 -> <missing>          <missing>          <missing>
2010-01-22 00:00:00 -> <missing>          <missing>          194.2937028157957
2010-01-25 00:00:00 -> <missing>          <missing>          <missing>
2010-01-26 00:00:00 -> <missing>          <missing>          <missing>
2010-01-27 00:00:00 -> <missing>          <missing>          <missing>
2010-01-28 00:00:00 -> <missing>          <missing>          <missing>
2010-01-29 00:00:00 -> <missing>          <missing>          <missing>

没有遗漏的数字是正确的。但是我想保留没有分红的价格

let dfZipped3 = priceFrame.Zip(divFrame, JoinKind.Left, JoinKind.Left, Lookup.Exact, false, fun (p:float) d -> p + (d |> Option.defaultValue 0.0))
        dfZipped3.Print()

结果

                       28881              29021              29399
2010-01-01 00:00:00 -> 123.535878499576   195.28635425580265 189.92210186152082
2010-01-04 00:00:00 -> 124.19087548338847 198.10448102247753 190.1571733631235
2010-01-05 00:00:00 -> 123.82028508465247 197.8259452373992  190.31388769752525
2010-01-06 00:00:00 -> 124.17363872065654 197.80956077945342 189.98478759528152
2010-01-07 00:00:00 -> 123.4583130672824  197.58017836821244 190.31388769752527
2010-01-08 00:00:00 -> 124.23396739021821 198.10448102247756 190.25120196376457
2010-01-11 00:00:00 -> 125.12166067091142 197.87509861123658 190.73701640041008
2010-01-12 00:00:00 -> 124.9234378994945  195.0569718445617  191.41088803833776
2010-01-13 00:00:00 -> 125.06133200134975 195.64681233060992 191.50491663897884
2010-01-14 00:00:00 -> 124.97514818769021 196.28580619049552 191.56760237273951
2010-01-15 00:00:00 -> 123.71686450826103 192.5829186947483  192.08475967626538
2010-01-18 00:00:00 -> 123.71686450826103 194.10667328370621 192.31983117786805
2010-01-19 00:00:00 -> 123.15666971947407 195.87619474185092 191.94371677530378
2010-01-20 00:00:00 -> 121.5622691667727  191.79646471335064 192.82131704795376
2010-01-21 00:00:00 -> 121.5450324040408  188.38849746062752 192.9937028157957
2010-01-22 00:00:00 -> 121.81220222638535 186.8647428716696  192.9937028157957
2010-01-25 00:00:00 -> 121.94147794687466 184.83307008639233 192.9937028157957
2010-01-26 00:00:00 -> 121.38990153945363 185.9799821425972  193.19743145051802
2010-01-27 00:00:00 -> 120.94174570842405 184.91499237612123 193.3541457849198
2010-01-28 00:00:00 -> 120.44187958919875 182.5392459739825  193.22877431739838
2010-01-29 00:00:00 -> 119.4938576389439  183.75169586197052 193.35414578491978

所有价格都在那里，但 none 的股息已被添加

let dfZipped4 = priceFrame.Zip(divFrame, JoinKind.Left, JoinKind.Left, Lookup.Exact, true, fun (p:float) d -> p + (d |> Option.defaultValue 0.0))
        dfZipped4.Print()

除了缺失值外什么都没有。

如何在股息对齐时将价格添加到股息中，否则价格保持不变？

更新

我已经为Frocha和zuzhu的每一个答案的执行计时了。 zyzhu 的第二个答案就目前而言并没有产生正确的结果。

对于每种技术连续运行 1000 次，我得到的典型时间为

frocha1: 572.974400
frocha2: 562.867600
zyzhu1: 1099.057100

frocha2 始终比 frocha1 稍快。 zyzhu1 总是比其他人慢。所以现在我接受 Frocha 的回答。

但是如果zyzhu2可以运行，它可能是最快的，因为它是最简单的。在这种情况下，我将更改已接受的答案。

Answer 1

我的方法没有考虑速度限制，如下： 1) 重命名列以便能够无错误地执行 Join 2）加入框架。 3）用零替换缺失值。 4) 对相应的列求和。 5) 删除股息栏。 6) 可选：如果不需要 "string" 转换，则将 priceFrame 的 col 名称更改为原始类型。

module Frame =
    //I usually add this handy function to the Frame module
    let mapReplaceCol col f frame =
        frame
        |> Frame.replaceCol col (Frame.mapRowValues f frame)

let priceFrame' = priceFrame |> Frame.mapColKeys string

//appends a "D" in the col key to eliminate col with same name
let dividends' =
    dividends
    |> Frame.mapColKeys (string >> (+) "D") 

let joinedFrame =
    priceFrame'
    |> Frame.join JoinKind.Right dividends'
    |> Frame.fillMissingWith 0.

(joinedFrame,priceFrame'.ColumnKeys |> List.ofSeq)
||> List.fold (fun acc elem ->
    acc|> Frame.mapReplaceCol elem (fun row ->
        row.GetAs<float>("D" + elem) + row.GetAs<float>(elem))
    |> Frame.dropCol ("D" + elem))

编辑

另一种使用 Zip 的方法。

//generate a dividends frame with the same rows of priceFrame
let dividends2 =
    (priceFrame,priceFrame.ColumnKeys |> List.ofSeq)
    ||> List.fold (fun acc elem -> acc|> Frame.dropCol elem) //empty frame
    |> Frame.join JoinKind.Outer dividends
    |> Frame.fillMissingWith 0.

(priceFrame,dividends2) ||> Frame.zip (fun (p : float) (d : float) -> p + d)

Answer 2

要使用zip，两个框架将逐列匹配并将两个系列相加。

在您的案例中，divFrame 的观测值少于 priceFrame。当两个系列没有相同数量的观察值并加在一起时，将缺少不匹配的结果。

我的解决方案是创建一个虚拟框架，使 divFrame 首先与 priceFrame 对齐。

let divFrame2 =
  let dummy =
    priceFrame.RowKeys
    |> Seq.collect(fun row -> divFrame.ColumnKeys |> Seq.map(fun col -> row, col, 0) )
    |> Frame.ofValues
  (dummy + divFrame).FillMissing(0.)
priceFrame + divFrame2

Answer 3

我有类似的问题，我的解决方案与zyzhu类似，但可以处理多个帧

let zipAll (dfs:Frame<_,_>[]) = 
    let outerKeys = dfs |> Array.collect (fun df -> df.RowKeys |> Array.ofSeq) |> Array.distinct |> Array.sort
    let dfsNew = 
        dfs
        |> Array.map ( Frame.indexRowsWith outerKeys >> Frame.mapRowValues (Series.fillMissingWith 0.) >> Frame.ofRows)
    Array.fold (Frame.zip (+)) (Array.head dfsNew) (Array.tail dfsNew)

[|priceFrame;dividends2|] |> zipAll

它可能很慢，但它可以处理多帧。

如何压缩 2 个 Deedle 帧并处理缺失值？

How to zip 2 Deedle frames and handle missing values?

f#

dataframe

deedle