This package provides transforms that are commonly used in statistics and machine learning. It was developed to address specific needs in feature engineering and works with general Tables.jl tables.

Past attempts to model transforms in Julia such as FeatureTransforms.jl served as inspiration for this package. We are happy to absorb any missing transform, and contributions are very welcome.

Features

• Transforms are revertible meaning that one can apply a transform and undo the transformation without having to do all the manual work keeping constants around.

• Pipelines can be easily constructed with clean syntax (f1 → f2 → f3) ⊔ (f4 → f5), and they are automatically revertible when the individual transforms are revertible.

• Branches of a pipeline and colwise transforms are run in parallel using multiple threads with the awesome Transducers.jl framework.

• Pipelines can be reapplied to unseen "test" data using the same cache (e.g. constants) fitted with "training" data. For example, a ZScore relies on "fitting" μ and σ once at training time.

Rationale

A common task in statistics and machine learning consists of transforming the variables of a problem to achieve better convergence or to apply methods that rely on multivariate Gaussian distributions. This process can be quite tedious to implement by hand and very error-prone. We provide a consistent and clean API to combine statistical transforms into pipelines.

Although most transforms discussed here come from the statistical domain, our long term vision is more ambitious. We aim to provide a complete user experience with fully-featured pipelines that include standardization of column names, imputation of missing data, and more.

Installation

Get the latest stable release with Julia's package manager:

] add TableTransforms

Usage

Below is a quick example with simple transforms:

using TableTransforms
using PairPlots

# example table from PairPlots.jl
N = 100_000
a = [2randn(N÷2) .+ 6; randn(N÷2)]
b = [3randn(N÷2); 2randn(N÷2)]
c = randn(N)
d = c .+ 0.6randn(N)
table = (;a, b, c, d)

# corner plot of original table
table |> corner

# convert to PCA scores
table |> PCA() |> corner

# convert to any Distributions.jl
table |> Quantile(Normal()) |> corner

Below is a more sophisticated example with a pipeline that has two parallel branches. The tables produced by these two branches are concatenated horizontally in the final table:

# create a transform pipeline
f1 = ZScore()
f2 = Scale()
f3 = Quantile()
f4 = Functional(cos)
f5 = Interquartile()
pipeline = (f1 → f2 → f3) ⊔ (f4 → f5)

# feed data into the pipeline
table |> pipeline |> corner

To revert a pipeline or single transform, use the apply and revert functions instead:

# apply transform and save cache to revert later
newtable, cache = apply(pipeline, table)

# perform additional modeling with newtable
# newtable = ...

# revert pipeline when done with modeling
original = revert(pipeline, newtable, cache)

Finally, it is sometimes useful to reapply a transform that was "fitted" with training data to unseen test data. In this case, the cache from a previous apply call is used:

# ZScore transform "fits" μ and σ using training data
newtable, cache = apply(ZScore(), traintable)

# we can reuse the same values of μ and σ with test data
newtable = reapply(ZScore(), testtable, cache)

Available transforms

Please check the docstrings for additional information.

Builtin

External

Custom transforms

It is easy to integrate custom transforms into existing pipelines. The new transform should be a subtype of Transform, and should implement apply. If the new transform isrevertible, then it should also implement revert.

Contributing

Contributions are very welcome. Please open an issue if you have questions.

Below are a few transforms that you can contribute:

• Standardization of column names
• Imputation of missing entries
• ...any transform that is useful

Related packages

• FeatureTransforms.jl has transforms, but they are not fully revertible. Some of their transforms such as MeanStdScaling are constructed for a specific table and cannot be inserted in the middle of a pipeline for example.
• AutoMLPipeline.jl relies on the Python stack via PyCall.jl. They provide pipelines with Julia's pipe |> operator and follow a more "Pythonic" interface. They do not support general Tables.jl.
• Impute.jl, Cleaner.jl, DataConvenience.jl all have a small set of transforms related to fixing column names as well as other basic transforms that we plan to absorb in the long term.
• DataFramesMeta.jl is a package to manipulate DataFrames.jl tables. It is not intended for statistical transforms such as PCA, Quantile, etc, which rely on complex interactions between the rows and columns of a table. The usage of macros in the package promotes one-shot scripts as opposed to general pipelines that can be passed around to different places in the program.
• Query.jl is a package to query IterableTables.jl. Similar to other alternatives above, the package is not intended for advanced statistical transforms.