File size: 8,122 Bytes
c711202 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 | """
Main Paper Figure: Accuracy vs Compute Comparison
Creates a full-width figure with two panels:
- (a) Discretization convergence: L2 error vs grid size N
- (b) Pareto frontier: L2 error vs wall time
Uses results from run.jl with multiple IC seeds to show error bars (std across seeds).
Usage:
julia --project=../.. plot_main_comparison.jl --results results/accuracy_vs_compute/results_*.csv
julia --project=../.. plot_main_comparison.jl --results results/accuracy_vs_compute/results_*.csv --output figures/
"""
using CSV
using DataFrames
using CairoMakie
using TuePlots
using Statistics
using ArgParse
# ==============================================================================
# Style Constants
# ==============================================================================
const COLORS = Dict(
"sparse_fvm" => colorant"#0072B2", # Blue
"sparse_collocation" => colorant"#D55E00", # Vermillion/orange
"classical_fvm" => colorant"#009E73" # Teal/green
)
const MARKERS = Dict(
"sparse_fvm" => :circle,
"sparse_collocation" => :diamond,
"classical_fvm" => :utriangle
)
const LABELS = Dict(
"sparse_fvm" => "GP-FVM",
"sparse_collocation" => "GP-Collocation",
"classical_fvm" => "Classical FVM"
)
# ==============================================================================
# Helper Functions
# ==============================================================================
"""
Add panel label (a), (b), etc. at top-center of axis.
"""
function add_panel_label!(ax, label; fontsize=8)
text!(ax, 0.5, 0.97, text=label, align=(:center, :top),
fontsize=fontsize, font=:bold, space=:relative)
end
"""
Aggregate results across IC seeds: compute mean and std for each (method, N).
Returns DataFrame with columns:
- method, N
- mean_error, std_error (for final_l2_error)
- mean_time, std_time (for time_s)
"""
function aggregate_by_method_N(df)
# Group by method and N, aggregate across IC families and seeds
grouped = groupby(df, [:method, :N])
result = combine(grouped,
:final_l2_error => mean => :mean_error,
:final_l2_error => std => :std_error,
:time_s => mean => :mean_time,
:time_s => std => :std_time,
nrow => :n_samples
)
# Sort by N for proper line drawing
sort!(result, [:method, :N])
return result
end
# ==============================================================================
# Main Figure: Convergence + Pareto
# ==============================================================================
"""
plot_main_comparison(csv_path; output_path, height)
Create main paper figure: (a) Convergence, (b) Pareto frontier.
Full-width, 2 panels side-by-side, with error bars across IC seeds.
"""
function plot_main_comparison(csv_path::String;
output_path::String="figures/accuracy_comparison.pdf",
height::Float64=0.7)
df = CSV.read(csv_path, DataFrame)
# Aggregate across IC seeds
agg = aggregate_by_method_N(df)
println("\nAggregated data:")
for method in unique(agg.method)
mdata = filter(r -> r.method == method, agg)
println(" $method: $(nrow(mdata)) grid sizes, $(mdata.n_samples[1]) samples each")
end
# TuePlots theme: full-width, 2 panels
# For 2 panels side-by-side, each is ~3.375" wide, so ratio=0.7 gives ~2.4" height
theme = Theme(
TuePlots.SETTINGS[:ICML];
font=true, fontsize=true, figsize=true,
single_column=false, nrows=1, ncols=2,
subplot_height_to_width_ratio=height
)
set_theme!(theme)
fig = Figure()
# Get unique N values for x-axis ticks
all_N = sort(unique(agg.N))
# ==========================================================================
# (a) Convergence: Error vs N
# ==========================================================================
ax1 = Axis(fig[1,1],
xlabel=L"Grid size $N$",
ylabel=L"Relative $L^2$ error",
xscale=log10, yscale=log10,
xticks=(all_N, string.(all_N)))
for method in ["sparse_fvm", "sparse_collocation", "classical_fvm"]
mdata = filter(r -> r.method == method, agg)
if nrow(mdata) == 0
continue
end
# Plot line with markers
scatterlines!(ax1, mdata.N, mdata.mean_error,
color=COLORS[method],
marker=MARKERS[method],
markersize=6,
label=LABELS[method])
# Add error bars (std across IC seeds)
# For log scale, we need asymmetric bars to avoid negative values
err_low = min.(mdata.std_error, mdata.mean_error * 0.9) # Clamp to 90% of mean
err_high = mdata.std_error
errorbars!(ax1, mdata.N, mdata.mean_error, err_low, err_high,
color=COLORS[method],
whiskerwidth=5,
linewidth=1)
end
add_panel_label!(ax1, "(a)")
# ==========================================================================
# (b) Pareto: Error vs Time
# ==========================================================================
# Nice tick values for wall time
time_ticks = [0.5, 1, 2, 5, 10, 20, 50]
ax2 = Axis(fig[1,2],
xlabel="Wall time (s)",
ylabel=L"Relative $L^2$ error",
xscale=log10, yscale=log10,
xticks=(time_ticks, string.(time_ticks)),
yticklabelsvisible=false)
for method in ["sparse_fvm", "sparse_collocation", "classical_fvm"]
mdata = filter(r -> r.method == method, agg)
if nrow(mdata) == 0
continue
end
# Plot line with markers
scatterlines!(ax2, mdata.mean_time, mdata.mean_error,
color=COLORS[method],
marker=MARKERS[method],
markersize=6,
label=LABELS[method])
# Add error bars (std across IC seeds)
# For log scale, we need asymmetric bars to avoid negative values
err_low = min.(mdata.std_error, mdata.mean_error * 0.9)
err_high = mdata.std_error
errorbars!(ax2, mdata.mean_time, mdata.mean_error, err_low, err_high,
color=COLORS[method],
whiskerwidth=5,
linewidth=1)
end
add_panel_label!(ax2, "(b)")
# Link y-axes
linkyaxes!(ax1, ax2)
# Single legend below
Legend(fig[2, 1:2], ax1, orientation=:horizontal, tellheight=true, framevisible=false)
# Save
mkpath(dirname(output_path))
save(output_path, fig, pt_per_unit=1)
# Also save PNG preview
png_path = replace(output_path, ".pdf" => ".png")
save(png_path, fig, px_per_unit=3)
set_theme!() # Reset theme
println("\nSaved: $output_path")
println("Saved: $png_path")
return fig
end
# ==============================================================================
# CLI
# ==============================================================================
function parse_commandline()
s = ArgParseSettings(description = "Generate main paper figure for accuracy vs compute comparison")
@add_arg_table! s begin
"--results", "-r"
help = "Path to results CSV file (from run.jl)"
arg_type = String
required = true
"--output", "-o"
help = "Output directory for figures"
arg_type = String
default = "figures"
"--height"
help = "Subplot height-to-width ratio (0.7 = slightly wide panels)"
arg_type = Float64
default = 0.7
end
return parse_args(s)
end
function main()
args = parse_commandline()
csv_path = args["results"]
output_dir = args["output"]
println("="^60)
println("Main Paper Figure: Accuracy vs Compute")
println("="^60)
println("Input: $csv_path")
println("Output: $output_dir")
mkpath(output_dir)
output_path = joinpath(output_dir, "accuracy_comparison.pdf")
plot_main_comparison(csv_path; output_path=output_path, height=args["height"])
println("\nDone!")
end
if abspath(PROGRAM_FILE) == @__FILE__
main()
end
|