From 472c5af51e47547085b1677edac9def3c639f870 Mon Sep 17 00:00:00 2001
From: Jiri Kalvoda <jirikalvoda@kam.mff.cuni.cz>
Date: Mon, 6 May 2024 20:07:01 +0200
Subject: [PATCH] =?UTF-8?q?prace:=20P=C5=99id=C3=A1n=C3=AD=20tabulek?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
prace/bakalarka/index.md | 57 ++++++++++++++++++++++++++++++++++++++++
1 file changed, 57 insertions(+)
diff --git a/prace/bakalarka/index.md b/prace/bakalarka/index.md
index 6208bce3e..05df0570c 100755
--- a/prace/bakalarka/index.md
+++ b/prace/bakalarka/index.md
@@ -1206,6 +1206,63 @@ Z toho pak můžeme vyslovit hypotézu, že $\delta_{\algo{sdp}}(n) \le 0.34$ pr
Z naměřených dat také můžeme usuzovat, že $\algo{sdp}$ je lepší než libovolný z jiných představených algoritmů.
+```python {redefine=tmp}
+from bakalarka import data_lib, g
+data = g.load_main_test()
+import math
+
+def row(*args):
+ return pf.TableRow(*(pf.TableCell(pf.Plain(*(i if isinstance(i, list) else parse_string(f"{i:0.3}") if isinstance(i, float) else parse_string(str(i))))) for i in args))
+
+def gen_alg(pipeline_name, algo_name, name_suffix, floatpage, add_note=False):
+ pipeline = data.pipelines[pipeline_name]
+ by_n = data_lib.group_by_n(pipeline)
+ rows = []
+ percentils = [10,50,90]
+ print_errors = any(i.error or i.data.get("broken", False) for i in pipeline)
+ for n in sorted(by_n.keys()):
+ d = by_n[n]
+ scores = [ i.score/n for i in d]
+ scores.sort()
+ l = len(scores)
+ avg = sum(scores) / l
+ variance = sum((avg-i)**2 for i in scores)/(l-1)
+ errors = len([None for i in d if i.error or i.data.get("broken", False)])
+ rows.append(row(
+ n,
+ l,
+ *([errors] if print_errors else []),
+ avg,
+ math.sqrt(variance),
+ *(scores[int((l-0.001)*perc/100)] for perc in percentils),
+ ))
+
+ table = pf.Table(pf.TableBody(*rows), head=pf.TableHead(row(
+ [pf.Math("n", format='InlineMath')],
+ "testů",
+ *(["chyb"] if print_errors else []),
+ [pf.Math("\\overline{\\delta_{\\algo{"+algo_name+"}}(n)}", format='InlineMath')],
+ [pf.Math("\\sqrt{\widehat{\\delta_{\\algo{"+algo_name+"}}(n)^2}}", format='InlineMath')],
+ *([pf.Math(f"{i} \%", format='InlineMath')] for i in percentils)
+ )))
+ note = element.content if add_note else []
+ return processor.transform([pf.Figure(table, *note, caption=pf.Caption(pf.Plain(*parse_string("Statistika algoritmu "), pf.Math(f"\\algo{{{algo_name}}}", format="InlineMath"), *name_suffix, *parse_string("."))), attributes=dict(floatpage=floatpage))])
+return [
+ *gen_alg("greedy", "g", [], "stat1", True),
+ *gen_alg("rg", "rg", [], "stat1!"),
+ *gen_alg("rsg", "rsg", [], "stat2"),
+ *gen_alg("semidef_prog_sage.sage(10, CVXOPT)", "sdp", parse_string(" – Sage"), "stat2"),
+ *gen_alg("semidef_prog(10)", "sdp", parse_string(" – SDPA-C"), "stat2!"),
+ ]
+```
+::: {c=tmp}
+Veškeré hodnoty jsou zaokrouhleny na 3 platné číslice.
+
+$\overline{\delta_{\alg}(n)}$ značí výběrový průměr, tedy $\frac{1}{m}\sum_{0\le i < m} r_i$, kde $m$ je počet testů a $r_i$ je relativní skóre $i$-tého z nich.
+
+$\widehat{\delta_{\alg}(n)^2}$ značí výběrový rozptyl, tedy $\frac{1}{m-1}\sum_{0\le i < m} \left(r_i - \overline{\delta_{\alg}(n)}\right)^2$.
+:::
+
\vfil\eject
Dolní odhad
--
GitLab