# Stack via flatmap
**URL:** https://discourse.julialang.org/t/stack-via-flatmap/130218
**Category:** General Usage
**Created:** [June 26, 2025, 3:29am UTC](https://discourse.julialang.org/t/stack-via-flatmap/130218 "2025-06-26T03:29:12Z")
**Posts on this page:** 5
**Page:** 1
### Author:  [@jar1](https://discourse.julialang.org/u/jar1)
#### Post date: [June 26, 2025, 3:29am UTC](https://discourse.julialang.org/t/stack-via-flatmap/130218/1 "2025-06-26T03:29:13Z")
```julia
using CSV, DataManipulation, StructArrays, Tables
julia> @p begin
"""
Year,CountryName,Population!!Estimate,Population!!MarginOfError,GDP!!Estimate,GDP!!MarginOfError
2025,China,1400000000,100000000,18000000000000,1000000000000
2025,India,1400000000,100000000,3000000000000,1000000000000
"""
StructArray(columntable(CSV.File(IOBuffer(__))))
flatmap() do r
[
(;r.Year, r.CountryName, k => getproperty(r, k))
for k in propertynames(r)[3:end]
]
end
end
8-element Vector{NamedTuple{names, Tuple{Int64, String7, Int64}} where names}:
(Year = 2025, CountryName = "China", Population!!Estimate = 1400000000)
(Year = 2025, CountryName = "China", Population!!MarginOfError = 100000000)
(Year = 2025, CountryName = "China", GDP!!Estimate = 18000000000000)
(Year = 2025, CountryName = "China", GDP!!MarginOfError = 1000000000000)
(Year = 2025, CountryName = "India", Population!!Estimate = 1400000000)
(Year = 2025, CountryName = "India", Population!!MarginOfError = 100000000)
(Year = 2025, CountryName = "India", GDP!!Estimate = 3000000000000)
(Year = 2025, CountryName = "India", GDP!!MarginOfError = 1000000000000)
```
This is `DataFrames.stack` implemented with `flatmap`. It works fine but I don’t like the code style.
- It uses column names for two and indexes for the rest. I’d rather (1) use the two names and refer to the rest by omission, or (2) say `1:2` and `3:end`.
- It requires `getproperty` - ugly.
- It requires a list comprehension - verbose.
Is there a nicer way to do this?
cc @aplavin
---
### Author:  [@aplavin](https://discourse.julialang.org/u/aplavin)
#### Post date: [June 26, 2025, 3:27pm UTC](https://discourse.julialang.org/t/stack-via-flatmap/130218/2 "2025-06-26T15:27:53Z")
Hmm, are you sure that’s the result you want? In your example, the last field name is different for every row, it will likely be inconvenient to work with that downstream.
For example, this one would select all columns with `!!` for flattening, put the original column names as `col` and values as `val`:
```julia
flatmap(pairs(_[sr".*!!.*"]), (;_.Year, _.CountryName, col=_2.first, val=_2.second))
```
Btw, for more convenient processing, you may want to immediately combine estimate + error into one Julian object.
And one minor thing: you can just pass `StructArray` to `CSV.read` directly.
---
### Author:  [@jar1](https://discourse.julialang.org/u/jar1)
#### Post date: [June 26, 2025, 4:55pm UTC](https://discourse.julialang.org/t/stack-via-flatmap/130218/3 "2025-06-26T16:55:28Z")
> [@aplavin](#):
>
> Hmm, are you sure that’s the result you want?
You’re right of course, I meant
```julia-auto
flatmap() do r
[(;r.Year, r.CountryName, :col => p, :val => getproperty(r, p))
for p in propertynames(r)[3:end]]
end
┌───────┬─────────────┬───────────────────────────┬────────────────┐
│ Year │ CountryName │ col │ val │
│ Int64 │ String7 │ Symbol │ Int64 │
├───────┼─────────────┼───────────────────────────┼────────────────┤
│ 2025 │ China │ Population!!Estimate │ 1400000000 │
│ 2025 │ China │ Population!!MarginOfError │ 100000000 │
│ 2025 │ China │ GDP!!Estimate │ 18000000000000 │
│ 2025 │ China │ GDP!!MarginOfError │ 1000000000000 │
│ 2025 │ India │ Population!!Estimate │ 1400000000 │
│ 2025 │ India │ Population!!MarginOfError │ 100000000 │
│ 2025 │ India │ GDP!!Estimate │ 3000000000000 │
│ 2025 │ India │ GDP!!MarginOfError │ 1000000000000 │
└───────┴─────────────┴───────────────────────────┴────────────────┘
```
or better
```julia-auto
flatmap() do r
[(;r.Year, r.CountryName, :col => k, :val => v)
for (k,v) in collect(pairs(r))[3:end]]
end
```
which solves #2 of my three gripes.
`_[sr".*!!.*"])` solves #1. And the 3-arg `flatmap` method solves #3.
Superb. Thank you!
---
### Author:  [@jar1](https://discourse.julialang.org/u/jar1)
#### Post date: [June 26, 2025, 5:06pm UTC](https://discourse.julialang.org/t/stack-via-flatmap/130218/4 "2025-06-26T17:06:20Z")
> [@aplavin](#):
>
> combine estimate + error into one Julian object.
Do you have in mind Measurements.jl `measurement(_, _)` or something else?
---
### Author:  [@aplavin](https://discourse.julialang.org/u/aplavin)
#### Post date: [June 30, 2025, 10:10am UTC](https://discourse.julialang.org/t/stack-via-flatmap/130218/5 "2025-06-30T10:10:03Z")
Any of the “numbers-with-uncertainties” packages really, there are a few and they have different usecases and tradeoffs. Basically a spectrum, I mostly use #1 or #3:
- MonteCarloMeasurements.jl for the most general error propagation with Monte Carlo samples
- Measurements.jl for linear error propagation with correlations (can be either faster or slower than the former)
- Uncertain.jl for linear error propagation without correlations, much less overhead than the first two