Skip to content

Add clustering using simulated annealing #215

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
bobot merged 3 commits into from
Aug 9, 2025
Merged

Add clustering using simulated annealing #215

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
66d6998
[Chart] clusterize using the correlation
bobot Aug 9, 2025
784d88e
keep legend near the graph
bobot Aug 9, 2025
d49f51a
[Chart] select with stroke instead of opacity for solvers
bobot Aug 9, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
166 changes: 116 additions & 50 deletions smtcomp/generate_graphics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,8 @@
from smtcomp.utils import *
import smtcomp.generate_website_page
import frontmatter
from random import Random
import math

c_file = pl.col("file")
c_logic = pl.col("logic")
Expand All @@ -28,6 +30,48 @@
c_bucket2 = pl.col("bucket2")


def correlation_sorting(solvers: List[U], corrs: Mapping[Tuple[U, U], float], nb_iteration: int) -> None:
"""
Strangely I can't find easy to use lib creating a well clustered correlation matrix. block modelling.

We use simulated annealing
"""
if len(solvers) <= 2:
return

r = Random(0)

def neighbor(i: int) -> float:
s = 0.0
n = 0
if 0 < i:
s += 1 - corrs[solvers[i - 1], solvers[i]]
n += 1
if i < len(solvers) - 1:
s += 1 - corrs[solvers[i], solvers[i + 1]]
n += 1
# At the bounds there is only one neighbor
return s / n

def swap(i: int, j: int) -> None:
tmp = solvers[i]
solvers[i] = solvers[j]
solvers[j] = tmp

for i in range(0, nb_iteration):
a = r.randint(0, len(solvers) - 1)
b = r.randint(0, len(solvers) - 2)
if a <= b:
b += 1
s1 = neighbor(a) + neighbor(b)
swap(a, b)
s2 = neighbor(a) + neighbor(b)
swap(a, b)
t = 1 - (i / (nb_iteration + 1))
if s2 < s1 or math.exp((s1 - s2) / t) > r.uniform(0.0, 1.0):
swap(a, b)


def create_output(
config: defs.Config,
results: pl.LazyFrame,
Expand Down Expand Up @@ -56,6 +100,7 @@ def create_output(
results.group_by(c_solver, c_solver2)
.agg(corr=pl.corr(c_cputime_s, c_cputime_s2, method="pearson"))
.sort(c_solver, c_solver2)
.select(c_solver, c_solver2, "corr")
)

results = (
Expand All @@ -78,39 +123,62 @@ def create_output(
).lazy()
results = results.join(lf_lookup, how="left", on=replaced).drop(replaced).rename({"pretty": replaced})

buckets = results.select(c_bucket.unique())

df_corr, df_results, df_buckets = pl.collect_all([corr, results, buckets])
df_corr, df_results, df_buckets, df_answers, df_solvers = pl.collect_all(
[
corr,
results,
results.select(c_bucket.unique()),
results.select(c_answer.unique()),
results.select(c_solver.unique()),
]
)

bucket_domain: list[float] = list(df_buckets["bucket"])
answer_domain: list[str] = list(df_answers["answer"])
answer_domain.sort(key=lambda x: defs.Answer(x).id)

solver_domain: list[str] = list(df_solvers["solver"])
solver_domain.sort(key=lambda x: x.lower())

row1 = df_corr.row(min(1, len(df_corr) - 1), named=True)
if True:
# Two provers can have no benchmars in common, their pairs is not in df_corrs
corrs: DefaultDict[Tuple[str, str], float] = defaultdict(lambda: 0.0)
for row in df_corr.rows(named=False):
corrs[row[0], row[1]] = row[2]
correlation_sorting(solver_domain, corrs, 1000)

# Create heatmap with selection
select_x = alt.selection_point(fields=["solver"], name="solver1", value=row1["solver"], toggle=False)
select_y = alt.selection_point(fields=["solver2"], name="solver2", value=row1["solver2"], toggle=False)
answer_x = alt.selection_point(fields=["answer"], name="answer1")
answer_y = alt.selection_point(fields=["answer2"], name="answer2")
solvers = alt.selection_point(
fields=["solver", "solver2"],
name="solvers",
value=[{"solver": solver_domain[0], "solver2": solver_domain[min(1, len(solver_domain) - 1)]}],
toggle=False,
)
answer_xy = alt.selection_point(fields=["answer", "answer2"], name="answer")
logic = alt.selection_point(fields=["logic"], name="logic")
division = alt.selection_point(fields=["division"], name="division")
g_select_provers = (
alt.Chart(df_corr, title="Click a tile to compare solvers", height=250, width=250)
.mark_rect()
.encode(
alt.X("solver", title=None),
alt.Y("solver2", title=None),
alt.X("solver", title="solver1").scale(domain=solver_domain),
alt.Y("solver2", title="solver2").scale(domain=list(reversed(solver_domain))),
alt.Color("corr", scale=alt.Scale(domain=[-1, 1], scheme="blueorange")),
opacity=alt.when(select_x, select_y).then(alt.value(1)).otherwise(alt.value(0.4)),
stroke=alt.when(solvers).then(alt.value("lightgreen")),
strokeWidth=alt.value(3),
opacity=alt.value(0.8),
)
.add_params(select_x, select_y)
.add_params(solvers)
)

title = alt.Title(alt.expr(f'{select_x.name}.solver + " vs " + {select_y.name}.solver2'))
g_results = (
# Number of results by time
title = alt.Title(
alt.expr(f'{solvers.name}.solver + " vs " + {solvers.name}.solver2'), subtitle="comparison of cpu time"
)
g_results_base = (
alt.Chart(df_results, title=title, height=250, width=250)
.transform_filter(select_x, select_y, answer_x, answer_y)
.transform_filter(solvers, answer_xy, logic, division)
.transform_aggregate(benchs="sum(len)", groupby=["bucket", "bucket2"])
.add_params(select_x, select_y, answer_x, answer_y)
.encode(
alt.X("bucket:O").axis(title="solver1", bandPosition=0.5)
# align should move the center of the cell, it does not
Expand All @@ -122,42 +190,40 @@ def create_output(
)
)

g_results_rect = g_results.mark_rect().encode(alt.Color("benchs:Q", scale=alt.Scale(scheme="yellowgreenblue")))

g_results_text = g_results.mark_text(baseline="middle")

# g_answers = (
# alt.Chart(df_results, title="Answers")
# .transform_filter(select_x, select_y, answer_x, answer_y)
# .transform_aggregate(benchs="sum(len)", groupby=["answer_x", "bucket2"])
# .add_params(select_x, select_y, answer_x, answer_y)
# .encode(
# alt.X("bucket:O").axis(title="solver1", bandPosition=0.5)
# # align should move the center of the cell, it does not
# .scale(type="band", align=0),
# alt.Y("bucket2:O").axis(title="solver2", bandPosition=0.5).scale(type="band", align=0).sort("descending"),
# text="benchs:Q",
# )
# .mark_bar()
# )
opacity = (
alt.when(select_x, select_y, answer_x, answer_y, logic, division).then(alt.value(1)).otherwise(alt.value(0.0))
)
g_results_rect = g_results_base.mark_rect().encode(alt.Color("benchs:Q", scale=alt.Scale(scheme="yellowgreenblue")))

legend_answer_x = (
alt.Chart(df_results, title=alt.Title(alt.expr(f'"solver1:" + {select_x.name}.solver')))
.mark_point()
.encode(alt.Y("answer:N").axis(title="", orient="right"), opacity=opacity)
.add_params(answer_x)
)
g_results_text = g_results_base.mark_text(baseline="middle")

legend_answer_y = (
alt.Chart(df_results, title=alt.Title(alt.expr(f'"solver2:" + {select_y.name}.solver2')))
.mark_point()
.encode(alt.Y("answer2:N").axis(title="", orient="right"), opacity=opacity)
.add_params(answer_y)
g_results = g_results_rect + g_results_text

opacity = alt.when(solvers, answer_xy, logic, division).then(alt.value(1)).otherwise(alt.value(0.01))

opacity_answer = alt.when(answer_xy).then(alt.value(1)).otherwise(alt.value(0.5))

# Number of results by answers
g_answer_base = (
alt.Chart(df_results, title="Comparison of the answers", height=250, width=250)
.transform_filter(solvers, logic, division)
.transform_aggregate(benchs="sum(len)", groupby=["answer", "answer2"])
.add_params(answer_xy)
.encode(
alt.X("answer:N").axis(title="solver1", bandPosition=0.5)
# align should move the center of the cell, it does not
.scale(type="band", align=0, domain=answer_domain),
alt.Y("answer2:N")
.axis(title="solver2", bandPosition=0.5)
.scale(type="band", align=0, domain=list(reversed(answer_domain))),
text="benchs:Q",
opacity=opacity_answer,
)
)

g_answer_rect = g_answer_base.mark_rect().encode(alt.Color("benchs:Q", scale=alt.Scale(scheme="yellowgreenblue")))

g_answer_text = g_answer_base.mark_text(baseline="middle")

g_answer = g_answer_rect + g_answer_text

legend_logic = (
alt.Chart(df_results, title="Logic")
.mark_point()
Expand All @@ -172,9 +238,9 @@ def create_output(
.add_params(division)
)

graph: alt.api.ChartType = (g_select_provers | g_results_rect + g_results_text).resolve_scale(color="independent")
graph: alt.api.ChartType = (g_select_provers | g_results).resolve_scale(color="independent")

graph = alt.vconcat(graph, legend_answer_x | legend_answer_y | legend_logic | legend_division)
graph = alt.vconcat(graph, (g_answer | (legend_logic | legend_division)).resolve_scale(color="independent"))

graph = graph.resolve_scale(color="independent")

Expand Down