From 052642b8aa04ddccbe684c65333c3a959736c479 Mon Sep 17 00:00:00 2001 From: Aria Shrimpton Date: Thu, 7 Mar 2024 16:18:56 +0000 Subject: new vis code --- analysis/vis.livemd | 42 ++++++++++++++++++++++++++---------------- 1 file changed, 26 insertions(+), 16 deletions(-) (limited to 'analysis') diff --git a/analysis/vis.livemd b/analysis/vis.livemd index f4b2ca6..727a459 100644 --- a/analysis/vis.livemd +++ b/analysis/vis.livemd @@ -6,7 +6,10 @@ Mix.install([ {:kino_vega_lite, "~> 0.1.8"}, {:json, "~> 1.4"}, {:explorer, "~> 0.8.0"}, - {:kino_explorer, "~> 0.1.11"} + {:kino_explorer, "~> 0.1.11"}, + {:nx, "~> 0.5"}, + {:scholar, "~> 0.2.1"}, + {:math, "~> 0.7.0"} ]) ``` @@ -17,7 +20,7 @@ require Explorer.DataFrame require Explorer.Series alias Explorer.DataFrame, as: DF alias Explorer.Series, as: SE -job_id = "1177" +job_id = "other" job_dir = Path.expand(~c"./" ++ job_id) |> Path.absname() sections_dir = Path.join(job_dir, "sections") cm_dir = Path.join([job_dir, "candelabra", "benchmark_results"]) @@ -93,9 +96,9 @@ cost_models ## Cost model exploratory plots ```elixir -startn = 0 -endn = 60_000 -resolution = 100 +startn = 200 +endn = 2000 +resolution = 50 points_for = fn impl, op -> %{"coeffs" => [coeffs]} = @@ -103,16 +106,18 @@ points_for = fn impl, op -> |> DF.to_columns() Enum.map(startn..endn//resolution, fn n -> + t = + (coeffs + |> Enum.take(3) + |> Enum.with_index() + |> Enum.map(fn {coeff, idx} -> coeff * n ** idx end) + |> Enum.sum()) + Enum.at(coeffs, 3) * Math.log2(n) + %{ impl: String.split(impl, "::") |> List.last(), op: op, n: n, - t: - coeffs - |> Enum.with_index() - |> Enum.map(fn {coeff, idx} -> coeff * n ** idx end) - |> Enum.sum() - |> max(0) + t: t } end) |> DF.new() @@ -123,6 +128,7 @@ end ```elixir inspect_op = "insert" +impls = ["BTreeSet", "EagerSortedVec", "EagerUniqueVec", "HashSet"] Tucan.layers([ cost_models @@ -131,19 +137,23 @@ Tucan.layers([ |> DF.to_rows() |> Enum.map(fn %{"impl" => impl} -> points_for.(impl, inspect_op) end) |> DF.concat_rows() - |> Tucan.lineplot("n", "t", color_by: "impl", clip: true) - |> Tucan.Scale.set_y_domain(0, 200), + |> DF.filter(impl in ^impls) + |> Tucan.lineplot("n", "t", color_by: "impl", clip: true), + # |> Tucan.Scale.set_y_domain(0, 200) Tucan.scatter( cost_model_points - |> DF.filter(op == ^inspect_op) - |> DF.group_by(["impl", "n"]) - |> DF.summarise(t: mean(t)), + |> DF.filter(op == ^inspect_op and impl in ^impls) + |> DF.group_by(["impl", "n"]), + # |> DF.summarise(t: mean(t)), "n", "t", color_by: "impl", clip: true ) ]) +|> Tucan.Scale.set_x_domain(startn, endn) +|> Tucan.Scale.set_y_domain(0, 200) +|> Tucan.set_size(500, 500) ``` ## Read benchmark data -- cgit v1.2.3