Release Notes

7.0.1 - 2026-04-26

Performance

• Stats moving/expanding windows: eliminate O(N) intermediate Series<'K, float> allocation in applyMovingSumsTransform and applyExpandingMomentsTransform. Values are now streamed lazily from the original series, reducing peak memory usage for large series.
• Stats updateSumsDense / initSumsDense: replace pown x 2/3/4 with explicit multiplications (x * x, x * x * x, x * x * x * x) in the per-step update function, eliminating the square-and-multiply loop overhead on each window step.

Infrastructure

• CI cache keys now include .config/dotnet-tools.json so tool version changes correctly invalidate the cached NuGet packages.

7.0.0 - 2026-04-20

Breaking changes

• Rename NuGet package Deedle.Excel.Reader → Deedle.ExcelReader to match the existing Deedle.ExcelReader namespace (closes #713)

New packages

• Deedle.ExcelWriter: cross-platform .xlsx writing via MiniExcel — Frame.writeExcel, Frame.writeExcelSheet, Frame.writeExcelSheets, stream overloads, C# ExcelFrameWriter static API (closes #714)

Bug fixes

• Deedle.Parquet: Fix Frame.writeParquet and Frame.writeParquetStream writing empty files (only column schema, no row data). Root cause: the IParquetRowGroupWriter returned by writer.CreateRowGroup() was not disposed, so the row group was never committed to the stream. Fixed by using use rg = writer.CreateRowGroup().

6.0.1 - 2026-04-13

Infrastructure

• Deedle.Arrow, Deedle.Parquet, and Deedle.ExcelReader are now F# namespaces instead of top-level modules. Functions like readArrow, readParquet, and readExcel are no longer brought into scope directly by open; use Frame.readArrow, Frame.readParquet, Frame.readExcel, etc. instead. C# static factory types (ArrowFrame, ParquetFrame, ExcelFrame) and extension methods are unchanged.
• Removed obsolete Deedle.PerfTest and Deedle.PerfTests projects (.NET Framework–only performance runner); use Deedle.Benchmarks (BenchmarkDotNet) instead
• Rename Deedle.Interactive package to Deedle.DotNetInteractive to match the folder name and clarify purpose
• Rename Deedle.Math package and namespace to Deedle.MathNetNumerics to clarify the MathNet.Numerics dependency; the old Deedle.Math namespace is preserved with [<Obsolete>] type aliases
• Remove stale docs and project solution folders from Deedle.sln
• Reorganise documentation into three categories: Guides, Integrations, Design

Documentation

• Add docs/dotnetinteractive.fsx — documentation page for Deedle.DotNetInteractive notebook formatting with configuration reference

New packages

• Deedle.Parquet: Apache Parquet interop — Frame read/write via Parquet.Net, stream I/O, row-key preservation, C# extensions

C# API improvements

• Add Frame.Stack / Frame.Unstack C# extension methods for reshaping frames between wide and long format
• Add Frame.MeltBy C# extension method for pandas-style wide-to-long pivot
• Add Frame.WhereRowsByMask / Frame.WhereColsByMask C# extension methods for boolean-mask filtering
• Add Frame.RenameColumn / Frame.RenameColumns C# extension methods for column renaming
• Add Frame.JoinOn / Frame.JoinOnString / Frame.JoinOnInt C# extension methods for joining frames on a column value
• Add Frame.Compare C# extension method for column-by-column frame diffing

5.0.0 - 2026-04-02

Documentation

• Add docs/excel.fsx — comprehensive documentation page for Deedle.Excel.Reader (cross-platform xlsx/xls reading) and Deedle.Excel (Windows live-Excel integration)

Performance

• Frame.stack: precompute per-column data vectors and per-row addresses before the nested loop, eliminating O(rows × cols) redundant index lookups
• Frame.unstack: single-pass extraction of unique r1 and r2 keys using HashSet-backed ResizeArray, replacing two separate Seq.map |> Seq.distinct |> Array.ofSeq traversals

New operations — Deedle.Math.Finance

• Add Finance.ewmCrossCov — EWMA pairwise covariance between two return series (mean-corrected, consistent with ewmCovMatrix) (#81)
• Add Finance.ewmCrossVol — signed square root of ewmCrossCov, giving cross-volatility in vol units with sign indicating direction of co-movement (#81)

Infrastructure

• Migrate to .NET 10 and FSharp.Core 10.0
• Update fsdocs-tool to 22.0.0-alpha.3
• Remove unnecessary System.Reflection.Emit package references (now in-box on net10.0)
• Suppress NU1510 warnings for in-box packages pulled transitively via Paket
• Suppress NU1701 warnings for NetOfficeFw packages (no net10.0 assets)
• Fix self-referential open FSharp.Data.Runtime in vendored TextRuntime.fs for newer F# compilers
• Replace deprecated Frame.indexRowsDate/Frame.indexRowsDateOffs usage in tests with Frame.indexRowsDateTime/Frame.indexRowsDateTimeOffset

Bug fixes

• Fix doc generation "No value returned by any evaluator" errors (#692)

4.0.1 - 2026-03-22

New packages

• Deedle.Arrow: Apache Arrow / Feather interop — Frame/Series ↔ RecordBatch, IPC I/O, Feather aliases, UInt/Date types, C# extensions, row-key preservation (#675, #681, #685)
• Deedle.MicrosoftML: ML.NET IDataView integration with VBuffer<float32>/VBuffer<float> vector column support (#677, #680)
• Deedle.Excel.Reader: cross-platform xlsx/xls reading via ExcelDataReader (#679)

New Frame operations

• Frame.interleave — side-by-side frame combining with tuple column keys (#691)
• Frame.joinOn, Frame.joinOnString, Frame.joinOnInt — join frames on a column value (#630, #663)
• Frame.nestRowsBy — hierarchical row nesting (#652)
• Frame.compare — column-by-column frame diffing (#637)
• Frame.rankRowsBy — rank rows within groups (#636)
• Frame.indexRowsApply — build row index from a projection (#650)
• Frame.indexRowsDateTime / Frame.indexRowsDateTimeOffset aliases (#653)
• Frame.mapColValuesAs — typed, non-boxing column mapping (#629)
• Frame.renameCol, Frame.renameColsUsing (#603)
• Frame.meltBy — pandas-style melt with identity columns (#598)
• Frame.distinctRowsBy — remove duplicate rows by column values (#596)
• Frame.toJson / Frame.saveJson — JSON serialisation (#609)
• Frame.pctChange — percentage change (#610)
• Frame.filterRowsByMask, Frame.filterColsByMask — boolean mask filtering (#619)
• Frame.empty / Series.empty module functions (#607)
• iloc integer-position indexing for Frame and Series (#631)

New Series operations

• Series.rank, Series.rankWith, Series.ntile — ranking and quantile binning (#636)
• Series.sweepLevel — hierarchical group normalisation (#638)
• Series.replaceValue — pandas-style value replacement (#628)
• Series.maskValues, Series.maskAll — replace values with missing (#605)
• Series.diffDate, Series.diffDateOffset — DateTime/DateTimeOffset differences (#592)
• Series.pctChange — percentage change (#610)
• Series.filterByMask — boolean mask filtering (#619)
• Series.before, Series.after, Series.startAt, Series.endAt, Series.between (#620)
• Series.windowWhileFromEnd, Series.chunkWhileFromEnd (#648)

New Stats operations

• Stats.movingMedian, Stats.expandingMedian (#661)
• Stats.levelMin, Stats.levelMax (#647)
• Stats.interpolateLinearWith — linear interpolation with configurable extrapolation (#627)
• Stats.cov, Stats.corr — pairwise series covariance and correlation (#614)
• Stats.corrMatrix, Stats.covMatrix — full correlation/covariance matrix for a Frame (#682)
• Stats.describe for Frame — pandas-style per-column summary statistics (#612)

New C# extension methods

• WindowWhile, ChunkWhile, PairwiseWith, MapRows, MapCols (#666)
• ZipInto, ZipAlignInto, ZipInner for Series (#657)

I/O enhancements

• Frame.ReadCsv accepts URLs (#658)
• Frame.ReadCsv typeResolver parameter for custom type inference (#649)
• Frame.ReadCsv / Frame.SaveCsv encoding parameter (#617)

Performance

• Eliminate seq/ref overhead in Stats moving and expanding window helpers (#683)
• Optimise Frame.AggregateRowsBy to avoid per-group sub-frame allocation (#669)
• Cache ToString() in CSV writer for repeated values (#621)

Bug fixes

• Fix delayed series display showing (Suppressed) — show key range instead (#662)
• Fix DateTime key formatting and locale-dependent formatting in FSI output (#633, #643)
• Fix Frame.fillMissingWith to handle numeric type widening (#632)
• Fix: throw InvalidOperationException for inexact Lookup on unordered series (#644)
• Fix DelayedSeries.Between to correctly update RangeMin/RangeMax (#624)
• Fix AsDecimal to support scientific notation (#623)
• Fix Frame.ReadCsv schema being ignored when inferTypes=false (#615)
• Fix Finance.ewmVol computing RMS instead of standard deviation (#593)
• Fix Frame.GetRowKeyAt to accept int instead of int64 (#588)
• Fix Frame.ofRecords for internal/private record types (#580)
• Fix IndexOutOfRangeException in Series.windowSize for series shorter than window (#573)
• Truncate long cell values in FSI/ToString output (#646)
• Fix filterRows/filterRowValues/Where losing ColumnTypes on all-missing columns (#601)
• Fix CSV schema parsing for column names containing parentheses (#604)
• Fix Frame.indexRowsWith to produce missing values for extra row keys (#595)

Documentation

• Add C# cookbook (docs/csharp.md) — covers SeriesBuilder, Frame.ReadCsv, Frame.FromRecords, column/row slicing via Columns[...]/Rows[...], GetColumn<T>, dynamic access, missing values, statistics, windowing, joining, and GetRowsAs<T> (#308)

Infrastructure

• Add SourceLink and deterministic build support (#660)
• Add BenchmarkDotNet benchmark suite under benchmarks/ (#687)
• Update test infrastructure: NUnit 4, FsUnit 7, FsCheck 3 (#616)
• Add [RequiresExplicitTypeArguments] to GetRow/TryGetRow and ObjectSeries typed accessors (#654)

4.0.0 - 2026-03-10

• Breaking change: Frame.stack and Frame.unstack now implement pandas-style reshape operations. stack converts Frame<'R,'C> to a long-format Frame<'R*'C, string> (tuple row keys, single "Value" column). unstack converts Frame<'R1*'R2,'C> to wide-format Frame<'R1, 'C*'R2> by promoting the inner row-key level to column keys.
• The old denormalised-table operations previously called stack/unstack were already renamed to melt/unmelt in an earlier release; the obsolete aliases are now removed.

4.0.0-alpha-001 - 2026-03-09

• Migrate to .NET 9
• Migrate documentation to fsdocs
• Remove FAKE build system, replace with simple build scripts
• Remove R Provider plugin

3.0.0 - 2023-01-17

• Fix missing value in Stats.cov552

3.0.0-beta.1 - 2022-06-22

• Add Deedle.DotNetInteractive package for formatting in dotnet notebooks
• Adjust and extend formatting interfaces accordingly.

2.5.0 - 2021-11-10

• Restore Deedle.RPlugin targeting net5.0

2.4.3 - 2021-09-11

• Add TryMin and TryMax in extension to be utilized in C#

2.4.2 - 2021-08-31

• Relax types for Matrix.dot of two frames 536

2.4.1 - 2021-07-25

• Add conversions between jagged array and frame 532

2.4.0 - 2021-07-06

• Add Frame.ReadCsvString 530
• .NET Standard 2.0 only

2.3.0 - 2020-09-16

• Add dropEmptyRows and dropEmptyCols 510
• Fix ambiguous sum function when calling it from Frame 515
• Update to latest NetOffice Excel package 505

2.2.0 - 2020-06-15

• Add linear regression and PCA functions in Deedle.Math thanks to @Ildhesten 496
• Add descriptions to Stats functions to clarify valid input types by @Arlofin 501

2.1.2 - 2020-02-25

• Fix parsing csv of multi-line column headers #479
• Fix type inference of empty cell #441

2.1.1 - 2019-11-4

• Fix FilterRowsBy #491
• Add basic stats functions of Frame in extension #490

2.1.0 - 2019-11-4

• Release Deedle.Math to extend linear algebra, statisitical analysis and financial analysis on Frame and Series by leveraging MathNet.Numerics #475
• Override + operator and add Frame.strConcat to concatenate Frame and Series of string values #482
• Fix ResampleUniform with missing values #470

2.0.4 - 2019-04-29

• Fix assembly version #472

2.0.3 - 2019-04-24

• Fix missing preferoptions in C# extension Frame.Readcsv from stream #471
• Optimize Frame.AggregateRowsBy #469

2.0.2 - 2019-04-08

• Fix missing signatures to control access of FSharp.Data implementations #465

2.0.1 - 2019-03-25

• Fix linear interpolation #458
• Fix FillMissingWith with nan #461
• Fix InvalidOperationException from ValuesAll #462
• Fix iterating seq multiple times in Frame.ofRecords #406

2.0.0 - 2019-03-13

• Breaking changes of Stats.min and Stats.max #422
• Inline stats functions #418
• Fix error handling of group by column when dealing with missing value 253 and 380
• Fix aggregateRowsBy with missing value 375
• Fix format function #416
• Fix arithmetic operator on frames #432
• Fix wrong exception for empty Series #365
• Fix tryLastValue using tryGetAt #339
• Fix bug in ReadCsv missingValues parameters #439
• Fix handling of missing column keys in case when inferTypes is false #63
• Add Frame.dropSparseRowsBy #404
• Add Series.intersect #407
• Add Series.compare #411
• Add Series.uniqueCount #413
• Add Series.describe #414 and $422
• Add Series.replace #427
• Add Stats.quantile for series #428
• Add Frame.slice #445
• Removed compiler warning from ReadCsv #426
• Rename stack/unstack to melt/unmelt #436

2.0.0-beta01 - 2018-08-04

• Support for netstandard2.0 #382, #393 and #391
• Excel support #255 and #399
• Iterate once in Frame.ofRowsOrdinal #396
• Fix for some concurrency errors #394
• Fix bug in Series.hasNot #361
• Fix bug in Frame.tryValues #359
• Arithmetic operators for decimal series #351

1.2.5

• Reading CSV (#332) and DropSparseRows (#333)
• Fix where filter in C# (#338)

1.2.4

• Fix RowsDense broken by BigDeedle changes (#319)
• Make ChunkSizeInto behave according to documentation (#314)
• Expand public fields (#313)
• Keep order of columns/rows in FrameBuilder (#322)

1.2.3

• Finish cleanup of BigDeedle code with partitioning support
• Add BigDeedle partitioning comment to design notes
• Update documentation tools dependencies

1.2.2

• BigDeedle: Materialize series on grouping and other operations
• BigDeedle: Support resampling without materializing series
• Better handling of materialization via addressing schemes
• Refactoring and cleanup of BigDeedle code
• Fix bugs in ordinal virtual index
• SelectOptional and SelectValues can be performed lazilly

1.2.1

• Support public fields in Frame.ofRecords

1.2.0

• Update version number for a BigDeedle release

1.1.5

• Aggregate bug fixes from previous beta releases
• Provide virtual index and virtual vector (aka BigDeedle)
• Compare indices using lazy sequences (to support BigDeedle)

1.1.4-beta

• Allow creation of empty ranges
• Support more operations on virtualized sources
• Fix handling of missing values in virtual Series.map

1.1.3-beta

• Introduce generic Ranges<T> type to simplify working with ranges (mainly useful for custom BigDeedle implementations)

1.1.2-beta

• Abstract handling of addresses (mainly for BigDeedle)
• Avoid accessing series Length in series and frame printing

1.1.1-beta

• Allow specifying custom NA values (#231)
• Documentation improvements and add F# Frame extension docs (#254)
• Use 100 rows for inference by default in C# and fix docs (#271)
• Fix R interop documentation issue (#287)
• More flexible conversion from R frames (#212)
• Dropping sparse rows/columns should preserve frame structure (#277)
• Change Stats.sum to return NaN for empty series (#259)
• Change C#-version of ReadCsv to accept inferTypes param (#270)

1.1.0-beta

• Enable materializing delayed series into a virtual series

1.0.7

• Add typed frame access (frame.GetRowsAs) (#281)
• BigDeedle improvements (#284, #285)
• Expose type information via frame.ColumnTypes (#286)
• Simplify load script (#292)
• Remove F# Data dependency & use Paket (#288, #293)
• Update depndencies (F# Formatting 2.6.2 and RProvider 1.1.8)

1.0.6

• Fix bugs related to frame with no columns (#272)
• Remove FSharp.Core dependency from BigDeedle public API

1.0.5

• Update R provider reference to 1.0.17

1.0.4

• Merge BigDeedle pull request (#247), add merging on big frames
• Fix PivotTable (#248) and CSV writing (#242)
• Update R provider reference to 1.0.16 (support shadow copy in F# 3.2.1)

1.0.3

• Added Stats.min and Stats.max for frame

1.0.2

• Operations GetAs, TryAs (ObjectSeries), GetColumns, GetRows, GetAllValues, ColumnApply (Frame) and filling of missing values uses "safe" conversion (allows conversion to bigger numeric type)
• Avoid boxing when filling missing values (#222)
• Fix documentation bugs (#221, #226) and update formatters from FsLab

1.0.1

• Update RProvider references

1.0.0

• Performance and API design improvements

1.0.0-alpha1

• API redesign, performance improvements and new features

1.0.0-alpha2

• Update to a new pre-release of RProvider

0.9.12

• Improved C# compatibility, added C# documentation

0.9.11-beta

• Fix bug when creating empty data frame

0.9.10-beta

• Support time series in the R plugin

0.9.9-beta

• Performance improvements, API additions, experimental R plugin

0.9.8-beta

• Add reflection-based frame expansion

0.9.7-beta

• Fix series formatting

0.9.6-beta

• Load script automatically references F# data (for CSV reading)

0.9.5-beta

• Update documentation and tools, adding functionality

0.9.4-beta

• Rename and various fixes and additions

0.9.3-beta

• Saving CSV, fix series alignment

0.9.2-beta

• Update paths in NuGet package

0.9.1-beta

• First beta version on NuGet

0.9.0-beta

• Initial release