# Learning to benchmark and find the best function to select a subset of a dataframe

**URL:** <https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704>\
**Category:** New to Julia\
**Tags:** question\
**Created:** [December 15, 2022, 1:46pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704 "2022-12-15T13:46:17Z")\
**Posts on this page:** 20\
**Page:** 1

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 15, 2022, 1:46pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/1 "2022-12-15T13:46:18Z")

</div>

Hello, I have issues related with the same piece of code:

```julia
using DataFrames
using BenchmarkTools

function fun1(df::DataFrame, bool_exp::BitVector)
    df_temp = df[bool_exp,:]
    return df_temp
end

function fun2(df::DataFrame, bool_exp::BitVector)
    df_temp = @subset(df, collect(bool_exp))
    return df_temp
end

function fun3!(df::DataFrame, bool_exp::BitVector)
    @subset!(df, collect(bool_exp))
    return nothing
end

df = DataFrame(x=rand(100),y=rand(100),z=rand(100))

bool_x = df.x .> 0.5

@time df1=fun1(df, bool_x)
@time df2=fun2(df, bool_x)
@time fun3!(df, bool_x)

```

The results are:

```julia
0.000015 seconds (12 allocations: 2.188 KiB)
0.043226 seconds (36.44 k allocations: 2.151 MiB, 97.86% compilation time)
0.044530 seconds (36.42 k allocations: 2.133 MiB, 98.24% compilation time)

```

I was expecting that the mutating function fun3! would be the optimal regarding time and memory. Why this is not true ?

The other question is how can I `@benchmark` a mutating function?  
Doing `@benchmark fun3!(df, bool_x)` gives error because the size of dataframe argument gets modified in each step of the process.

Finally, I would like to know if there are other better options to filter a dataframe that I haven’t considered.

Thank you very much!

---

<div class="post-metadata">

**Author:** ![lmiq](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lmiq/32/18314_2.png) [@lmiq](https://discourse.julialang.org/u/lmiq)\
**Post date:** [December 15, 2022, 2:19pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/2 "2022-12-15T14:19:48Z")

</div>

Use

```julia
@benchmark f(x) setup=(x=copy(x_input)) evals=1

```

where `x_input` is the initial value you want.

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 15, 2022, 2:58pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/3 "2022-12-15T14:58:52Z")

</div>

Thanks @lmiq !  
Doing as you say:

```julia
@benchmark df1=fun1(x, bool_x) setup=(x=copy(df)) evals=1
@benchmark df2=fun2(x, bool_x) setup=(x=copy(df)) evals=1
@benchmark fun3!(x, bool_x) setup=(x=copy(df)) evals=1

```

I obtain the following result:

```julia
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max): 840.000 ns … 30.324 μs ┊ GC (min … max): 0.00% … 0.00%
 Time (median): 1.282 μs ┊ GC (median): 0.00%
 Time (mean ± σ): 1.631 μs ± 1.233 μs ┊ GC (mean ± σ): 0.00% ± 0.00%

  ▇█▆▄▃▂▁▁▂▄▆▇▆▅▄▃▁▁▁▂▂▂▂▂▁ ▂
  ███████████████████████████▇▇▆▆▅▅▇▆▆▆▅▄▃▅▃▅▁▄▃▃▅▃▄▁▄▃▃▅▄▃▄▄▆ █
  840 ns Histogram: log(frequency) by time 6.46 μs <

 Memory estimate: 2.69 KiB, allocs estimate: 12.

BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max): 26.221 μs … 113.102 μs ┊ GC (min … max): 0.00% … 0.00%
 Time (median): 28.308 μs ┊ GC (median): 0.00%
 Time (mean ± σ): 29.833 μs ± 5.726 μs ┊ GC (mean ± σ): 0.00% ± 0.00%

  ▄████▇▆▅▄▃▃▂▁▁▁ ▁ ▂
  ████████████████▇▇▇▆▇▆▆▅▂▅████████▇▇▇▇▇▆▇▄▂▆▆▅▅▄▅▅▅▅▄▆▆▆▆▄▅▅ █
  26.2 μs Histogram: log(frequency) by time 57.7 μs <

 Memory estimate: 11.98 KiB, allocs estimate: 190.

BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max): 26.190 μs … 130.973 μs ┊ GC (min … max): 0.00% … 0.00%
 Time (median): 27.507 μs ┊ GC (median): 0.00%
 Time (mean ± σ): 28.823 μs ± 4.829 μs ┊ GC (mean ± σ): 0.00% ± 0.00%

  ▄▇█▇▆▅▄▃▃▂▁▁ ▂
  ██████████████▇▇▆▅▆▄▆▅▄▃▄▃▃▄▇█▇▇███▆▆▆▅▅▅▅▅▅▄▄▃▄▅▅▄▄▅▃▄▃▄▄▅▅ █
  26.2 μs Histogram: log(frequency) by time 54.3 μs <

 Memory estimate: 9.81 KiB, allocs estimate: 181.

```

Do you know why the first function is the best in terms of speed and memory usage? I was expecting that the mutating function to be the best. Are there other options?

---

<div class="post-metadata">

**Author:** ![lmiq](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lmiq/32/18314_2.png) [@lmiq](https://discourse.julialang.org/u/lmiq)\
**Post date:** [December 15, 2022, 3:58pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/4 "2022-12-15T15:58:41Z")

</div>

I know nothing about the `@subset` macros, but the `collect(bool_vector)` you are using in both other cases by itself is allocating an intermediate array which I guess is not necessary. I don’t see why you are collecting it there, as afaik collecting it just returns a copy of the same vector in this case.

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 15, 2022, 4:07pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/5 "2022-12-15T16:07:43Z")

</div>

Thanks. I use `collect` because it transforms the BitVector to Array{Bool}. I do not know why `@subset` needs the second object as argument.

---

<div class="post-metadata">

**Author:** ![pdeffebach](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pdeffebach/32/10320_2.png) [@pdeffebach](https://discourse.julialang.org/u/pdeffebach)\
**Post date:** [December 15, 2022, 4:31pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/6 "2022-12-15T16:31:32Z")

</div>

The need for `collect` is a bug. I just listed it and am tracking it [here](https://github.com/JuliaData/DataFramesMeta.jl/issues/348)

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 15, 2022, 4:39pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/7 "2022-12-15T16:39:01Z")

</div>

Thanks @pdeffebach.  
Once this bug is solved, using the mutating `@subset!` will be the optimum way to filter a dataframe, or is still the above `fun1` function the best ?

---

<div class="post-metadata">

**Author:** ![pdeffebach](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pdeffebach/32/10320_2.png) [@pdeffebach](https://discourse.julialang.org/u/pdeffebach)\
**Post date:** [December 15, 2022, 4:41pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/8 "2022-12-15T16:41:37Z")

</div>

It’s just a parsing issue. So if you re-run your benchmark with `identity` instead of `collect` you can benchmark.

---

<div class="post-metadata">

**Author:** ![pdeffebach](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pdeffebach/32/10320_2.png) [@pdeffebach](https://discourse.julialang.org/u/pdeffebach)\
**Post date:** [December 15, 2022, 4:43pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/9 "2022-12-15T16:43:33Z")

</div>

`fun1` is the best because there is a little overhead in the `DataFrames.subset` function that `@subset` calls. This overhead matters because your data frame is so small. Try doing with a million observations and see what happens.

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 15, 2022, 5:02pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/10 "2022-12-15T17:02:36Z")

</div>

Thanks!  
For dataframe with a million rows I obtain the following:

```julia
BenchmarkTools.Trial: 829 samples with 1 evaluation.
 Range (min … max): 2.934 ms … 12.817 ms ┊ GC (min … max): 0.00% … 59.91%
 Time (median): 3.368 ms ┊ GC (median): 0.00%
 Time (mean ± σ): 3.739 ms ± 1.274 ms ┊ GC (mean ± σ): 5.86% ± 11.59%

  ▁▃██▄▃▃▃▂ ▁ ▁                                               
  █████████████▇▇▄▆▁▅▅▄▁▄▅▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▇▇▆▆▁▄▅ █
  2.93 ms Histogram: log(frequency) by time 10.4 ms <

 Memory estimate: 15.23 MiB, allocs estimate: 20.

BenchmarkTools.Trial: 726 samples with 1 evaluation.
 Range (min … max): 3.878 ms … 11.137 ms ┊ GC (min … max): 0.00% … 57.65%
 Time (median): 4.291 ms ┊ GC (median): 0.00%
 Time (mean ± σ): 4.626 ms ± 1.206 ms ┊ GC (mean ± σ): 4.57% ± 10.52%

   ▂██▆▃▂▂▂ ▁▁                                               
  ██████████▇██▇▇▆▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▅▆▆█ ▇
  3.88 ms Histogram: log(frequency) by time 11 ms <

 Memory estimate: 15.36 MiB, allocs estimate: 201.

BenchmarkTools.Trial: 350 samples with 1 evaluation.
 Range (min … max): 11.574 ms … 19.007 ms ┊ GC (min … max): 0.00% … 38.25%
 Time (median): 11.887 ms ┊ GC (median): 0.00%
 Time (mean ± σ): 11.975 ms ± 642.910 μs ┊ GC (mean ± σ): 0.48% ± 3.33%

         ▄▂▄▄▄█▄▄▄▆▂▃ ▃                                       
  ▃▆▆▆▆███████████████▅▆█▄▄▄▄▅▃▇▃▃▃▄▃▃▃▃▃▃▁▁▁▁▁▁▃▁▁▁▁▃▁▁▁▃▁▁▁▃ ▄
  11.6 ms Histogram: frequency by time 12.9 ms <

 Memory estimate: 4.08 MiB, allocs estimate: 196.

```

`fun1` is faster but` fun3!` uses less memory. Correct? What is the variable “allocs estimate”? Please let me know which function would you use. Thanks!

---

<div class="post-metadata">

**Author:** ![pdeffebach](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pdeffebach/32/10320_2.png) [@pdeffebach](https://discourse.julialang.org/u/pdeffebach)\
**Post date:** [December 15, 2022, 5:07pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/11 "2022-12-15T17:07:46Z")

</div>

Huh. I dunno. I wonder if something is going on with the benchmarking. Maybe @bkamins has some intuition for why `fun3!` is so slow.

---

<div class="post-metadata">

**Author:** ![bkamins](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/bkamins/32/208538_2.png) [@bkamins](https://discourse.julialang.org/u/bkamins)\
**Post date:** [December 15, 2022, 5:52pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/12 "2022-12-15T17:52:51Z")

</div>

> [@pdeffebach](#):
>
> Maybe @bkamins has some intuition for why `fun3!` is so slow.

Yes, I can answer it:

```julia
julia> x = rand(10^6);

julia> bx = [v < 0.5 for v in x];

julia> ix = findall(!, bx);

julia> @benchmark z[bx] setup=(z=copy(x)) evals=1
BenchmarkTools.Trial: 745 samples with 1 evaluation.
 Range (min … max): 4.486 ms … 28.198 ms ┊ GC (min … max): 0.00% … 81.56%
 Time (median): 4.569 ms ┊ GC (median): 0.00%
 Time (mean ± σ): 4.939 ms ± 2.251 ms ┊ GC (mean ± σ): 4.37% ± 8.00%

  █▇▄▄▃▂▁▂▁▁
  ██████████▇▇▇█▇▇▆▅▆▆▁▅▅▅▆▇▆▆▄▇▅▅▇▆▄▁▄▁▁▁▄▁▁▁▁▁▁▁▁▁▄▁▁▁▁▁▁▄ ▇
  4.49 ms Histogram: log(frequency) by time 7.24 ms <

 Memory estimate: 3.82 MiB, allocs estimate: 2.

julia> @benchmark z[ix] setup=(z=copy(x)) evals=1
BenchmarkTools.Trial: 1398 samples with 1 evaluation.
 Range (min … max): 1.287 ms … 29.172 ms ┊ GC (min … max): 0.00% … 94.65%
 Time (median): 1.355 ms ┊ GC (median): 0.00%
 Time (mean ± σ): 1.699 ms ± 2.259 ms ┊ GC (mean ± σ): 12.15% ± 8.70%

  ██▅▄▂▁▁ ▁
  ███████▇▇█▇▆▄▇▅▄▄▄▄▄▁▄▄▄▄▄▆█▇▇▇▆▄▄▄▄▅▆▅▄▄▁▁▄▄▄▁▁▄▁▁▁▁▁▁▁▁▄ █
  1.29 ms Histogram: log(frequency) by time 4.27 ms <

 Memory estimate: 3.81 MiB, allocs estimate: 2.

julia> @benchmark deleteat!(z, bx) setup=(z=copy(x)) evals=1
BenchmarkTools.Trial: 1764 samples with 1 evaluation.
 Range (min … max): 798.600 μs … 3.617 ms ┊ GC (min … max): 0.00% … 0.00%
 Time (median): 825.700 μs ┊ GC (median): 0.00%
 Time (mean ± σ): 927.020 μs ± 304.150 μs ┊ GC (mean ± σ): 0.00% ± 0.00%

  ██▄▂▂▂▁ ▁
  ██████████▇▇▇▆▅▅▅▇▅▃▃▅▅▇▁▄▅▃▁▄▁▃▁▃▁▁▁▁▄▃▁▁▁▁▁▃▃▁▄██▇▇▇▆▁▄▁▄▃▅ █
  799 μs Histogram: log(frequency) by time 2.22 ms <

 Memory estimate: 0 bytes, allocs estimate: 0.

julia> @benchmark deleteat!(z, ix) setup=(z=copy(x)) evals=1
BenchmarkTools.Trial: 859 samples with 1 evaluation.
 Range (min … max): 3.794 ms … 6.477 ms ┊ GC (min … max): 0.00% … 0.00%
 Time (median): 3.828 ms ┊ GC (median): 0.00%
 Time (mean ± σ): 3.976 ms ± 316.706 μs ┊ GC (mean ± σ): 0.00% ± 0.00%

  █▆▂▂▂▁▃▁ ▁
  ████████████▇██▇▆▅▆▇▅▅▇▆▇▄▅▆▅▆▅▆▆▅▁▅▄▄▅▅▄▅▅▄▄▆▅▁▅▄▁▅▅▅▅▁▁▁▅ ▇
  3.79 ms Histogram: log(frequency) by time 5.21 ms <

 Memory estimate: 0 bytes, allocs estimate: 0.

```

And the current code in DataFrames.jl assumed that `deleteat!` and `getindex` had the same relationship in performance. Which clearly it has not. I will make a PR to fix it.

---

<div class="post-metadata">

**Author:** ![bkamins](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/bkamins/32/208538_2.png) [@bkamins](https://discourse.julialang.org/u/bkamins)\
**Post date:** [December 15, 2022, 7:27pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/13 "2022-12-15T19:27:38Z")

</div>

fixed in [fix deleteat! and subset! performance by bkamins · Pull Request #3249 · JuliaData/DataFrames.jl · GitHub](https://github.com/JuliaData/DataFrames.jl/pull/3249)

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 16, 2022, 5:45pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/14 "2022-12-16T17:45:49Z")

</div>

Hi @bkamins. As I am new to julia, I would like to clarify a few things.  
Could you explain me which is more performant:` deleteat!` or `getindex` ?  
Finally, I would like to know if updating DataFrames.jl includes this fix and which function do you suggest me to use `fun1, fun2` or the mutating `fun3!` (defined at the top of the thread) ?  
Best

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 16, 2022, 5:51pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/15 "2022-12-16T17:51:15Z")

</div>

I looked at the PR and saw you removed the use of the `findall` function. What are you using instead of it? What is the relation with `deleteat!` ?  
Thanks!

---

<div class="post-metadata">

**Author:** ![bkamins](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/bkamins/32/208538_2.png) [@bkamins](https://discourse.julialang.org/u/bkamins)\
**Post date:** [December 16, 2022, 6:16pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/16 "2022-12-16T18:16:11Z")

</div>

The answer to these questions is in my post above. The speed of operations is as follows (from fastest to slowest):

- `deleteat!` with `Bool` index
- `getindex` with integer index
- `deleteat!` with integer index
- `getindex` with `Bool` index

Note that `getindex` allocates a new vector, while `deleteat!` is in place. In the PR I remove `findall` as `findall` changes `Bool` index into integer index. This is desired for `getindex`, but bad for `deleteat!`. The original code for `deleteat!` was adapted from `getindex`, so `findall` stayed there, but should have been removed.

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 16, 2022, 6:33pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/17 "2022-12-16T18:33:39Z")

</div>

Thanks, I understood the ranking.  
I am trying to understand what does the `!` does in this expression:

> [@bkamins](#):
>
> `ix = findall(!, bx)`

Summing up, once I update the DataFrames package the best option will be to use my mutating `fun3!` function instead of the others?

---

<div class="post-metadata">

**Author:** ![bkamins](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/bkamins/32/208538_2.png) [@bkamins](https://discourse.julialang.org/u/bkamins)\
**Post date:** [December 16, 2022, 6:42pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/18 "2022-12-16T18:42:48Z")

</div>

`findall(!, bx)` means: find indices of all values in `bx` which are `true` after function `!` is applied to them. Since `!` is negation this means that we find indices of `false` values.

> once I update the DataFrames package

Precisely - after the PR is merged and a patch release is made. Since this is not a bug but performance issue the release might not be within a few days (we release bug fixes on daily basis).

If you want an in-place function then in the long run `fun3!` will be fastest, but it might take some time till the patch is released as I commented.

---

<div class="post-metadata">

**Author:** ![martinmestre](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/martinmestre/32/215857_2.png) [@martinmestre](https://discourse.julialang.org/u/martinmestre)\
**Post date:** [December 16, 2022, 6:53pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/19 "2022-12-16T18:53:44Z")

</div>

So `bx` is a boolean vector for those components`v<0.5`, while `ix` gives the indexes of the elements `v>0.5`. Isn’t the operations `z[bx]` and `z[ix]` doing opposite things? The same happens with `deleteat!(z,bx)` and `deleteat!(z,ix`) I guess? Thanks for the patience.

---

<div class="post-metadata">

**Author:** ![bkamins](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/bkamins/32/208538_2.png) [@bkamins](https://discourse.julialang.org/u/bkamins)\
**Post date:** [December 16, 2022, 7:01pm UTC](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704/20 "2022-12-16T19:01:40Z")

</div>

Yes, also the difference is that `z[ix]` and `deleteat!(z, ix)` do a different thing (one keeps indices, the other drops them). But since the threshold is 0.5 and we have 10^6 elements the number of elements to keep and to drop is roughly equal so it is not a problem.

I used `ix = findall(!, bx)` to make `z[ix]` and `deleteat!(z, bx)` equivalent as most likely you want the result of the operations to be the same and they are the performant options (but indeed I could have done it more carefully).

[Next page](https://discourse.julialang.org/t/learning-to-benchmark-and-find-the-best-function-to-select-a-subset-of-a-dataframe/91704.md?page=2)