this repo has no description
Continue rapport
+220
-31
+19
-25
autopush.ps1
+19
-25
autopush.ps1
···
1
1
#!/usr/bin/env pwsh
2
-
# PowerShell version of the Fish script using `mutool` for PDF page count
2
+
# PowerShell equivalent of the provided Fish script
3
3
4
-
# Checkout the PDF from Git
5
4
git checkout -- rapport/main.pdf
6
-
7
-
# Pull latest changes with rebase and autostash
8
5
git pull --rebase --autostash
9
6
10
7
while ($true) {
11
-
# --- Get the page count using mutool ---
12
-
try {
13
-
$pages = mutool info rapport/main.pdf |
14
-
Select-String "^Pages:" |
15
-
ForEach-Object { ($_ -split "\s+")[1] }
16
-
} catch {
17
-
Write-Host "Error: Could not get page count from rapport/main.pdf"
18
-
$pages = "0"
19
-
}
8
+
# Get page count using mutool instead of pdfinfo
9
+
$pageCount = & mutool info rapport/main.pdf |
10
+
Select-String '^Pages:' |
11
+
ForEach-Object { ($_ -split '\s+')[-1] }
20
12
21
-
# Save page count to a file
22
-
Set-Content -Path "pages_count" -Value $pages
13
+
$pageCount > pages_count
23
14
24
-
# --- Stage files ---
25
-
git add rapport/*.typ bib.yaml *.ps1 rapport/*.dot pages_count rapport/*.png
15
+
# Check for changes in relevant files
16
+
git diff --no-patch --exit-code slides/*.typ slides/*.dot slides/*.png rapport/*.typ rapport/*.dot rapport/*.png bib.yaml | Out-Null
17
+
$pdfChanges = $LASTEXITCODE
26
18
27
-
# --- Commit quietly ---
28
-
git commit --quiet -m "Continue rapport" 2>$null
19
+
Write-Host "PDF updates with these changes: $pdfChanges"
29
20
30
-
# --- Push quietly and force ---
31
-
git push --quiet --force
21
+
# Stage relevant files
22
+
git add rapport/ slides/ bib.yaml *.fish *.ps1 pages_count
23
+
git commit --quiet -m "Continue rapport"
32
24
33
-
# --- Print timestamp ---
34
-
Write-Host "Pushed at $(Get-Date -Format 'yyyy-MM-dd HH:mm:ss')"
25
+
# Push if there were changes
26
+
if ($pdfChanges -ne 0) {
27
+
git push --quiet --force
28
+
Write-Host "Pushed at $(Get-Date)"
29
+
}
35
30
36
-
# --- Sleep for 30 minutes ---
37
-
Start-Sleep -Seconds 1800
31
+
Start-Sleep -Seconds (60 * 30)
38
32
}
+5
-4
rapport/context.typ
+5
-4
rapport/context.typ
···
1
-
#import "utils.typ": comment, refneeded, todo
1
+
#import "utils.typ": comment, refneeded, todo, dontbreak
2
2
#import "@preview/fletcher:0.5.8": edge, node
3
3
#import "@preview/fletcher:0.5.8"
4
4
#import "@preview/diagraph:0.3.6"
···
8
8
caption: caption,
9
9
fletcher.diagram(..args),
10
10
)
11
-
#let dontbreak = content => block(breakable: false, content)
12
11
13
12
#show math.equation.where(block: true): set block(spacing: 2em)
14
13
···
252
251
253
252
On définit également l'ensemble de _tout_ les chemins d'états possibles, peut importe la politique, $cal(C)$ :
254
253
255
-
$
256
-
cal(C) :=
254
+
#let definitions_paths_set = $
255
+
cal(C) &:=
257
256
setbuilder(
258
257
cases(
259
258
& c_0 & = (s_0, a_0),
···
262
261
(s_0, a) in S times A^NN
263
262
)
264
263
$
264
+
265
+
#definitions_paths_set
265
266
266
267
On notera que, selon $M$, on peut avoir $cal(C) subset.neq (S times A)^NN$: par exemple, certains états de l'environnement peuvent représenter des "impasses", où il est impossible d'évoluer vers un autre état, peut importe l'action choisie.
267
268
slides/main.pdf
slides/main.pdf
This is a binary file and will not be displayed.
+195
-1
slides/main.typ
+195
-1
slides/main.typ
···
1
-
#import "../rapport/utils.typ": dontbreak
1
+
#import "../rapport/utils.typ": dontbreak, todo
2
+
#import "../rapport/context.typ": definitions_paths_set, exp
2
3
#import "@preview/touying:0.6.1": *
3
4
#import themes.simple: *
4
5
6
+
#import "@preview/fletcher:0.5.8": edge, node
7
+
#import "@preview/fletcher:0.5.8"
8
+
#import "@preview/diagraph:0.3.6"
9
+
10
+
#show figure: set block(spacing: 4em)
11
+
#let diagram = (caption: none, ..args) => figure(
12
+
caption: caption,
13
+
touying-reducer.with(reduce: fletcher.diagram, cover: fletcher.hide)(..args),
14
+
)
15
+
16
+
#let centered = content => {
17
+
v(1fr)
18
+
align(center, content)
19
+
v(1fr)
20
+
}
21
+
5
22
#show: simple-theme.with(aspect-ratio: "16-9")
6
23
#set text(font: "New Computer Modern")
7
24
#show raw: set text(font: "Martian Mono", size: 0.8em)
···
10
27
11
28
Gwenn Le Bihan `<gwenn.lebihan@etu.inp-n7.fr>` \
12
29
#datetime.today().display("[day padding:none] Novembre [year]") \
30
+
31
+
#title-slide[
32
+
== Reinforcement Learning
33
+
34
+
Et son application à la robotique
35
+
]
36
+
37
+
38
+
== Bases du RL
39
+
40
+
#diagram(
41
+
node((0, 0))[Agent],
42
+
node((1, 0))[Environnement],
43
+
node((2, 0))[Score],
44
+
45
+
pause,
46
+
edge((0, 0), (1, 0), "->")[Action],
47
+
48
+
pause,
49
+
edge((1, 0), (2, 0), "-->")[Fonction coût],
50
+
51
+
pause,
52
+
edge((2, 0), (0, 0), "->", bend: 45deg)[Mise à jour],
53
+
)
54
+
55
+
== RL en robotique
56
+
57
+
#diagram(
58
+
node((0, 0), todo[Photo H1v2]),
59
+
node((1, 0), todo[H1v2 dans gz]),
60
+
node((2, 0))[Score],
61
+
edge((0, 0), (1, 0), "->")[genou gauche +0.5°],
62
+
edge((1, 0), (2, 0), "-->", $cal(L)$),
63
+
edge((2, 0), (0, 0), "->", bend: 45deg)[Mise à jour],
64
+
)
65
+
66
+
67
+
68
+
#let what_is_surrogate_adv = [
69
+
== C'est quoi $cal(L)$ ?
70
+
71
+
#centered[
72
+
73
+
C'est très simple:
74
+
75
+
$
76
+
cal(L)_r (pi', pi) := exp_((s_t, a_t)_(t in NN) in cal(C)) sum_(t=0)^oo (Q_pi (s_t, a_t)) / (Q_pi' (s_t, a_t)) A_(pi, r)(s_t, a_t)
77
+
$
78
+
79
+
]
80
+
]
81
+
82
+
#what_is_surrogate_adv
83
+
84
+
#title-slide[
85
+
86
+
== Comparaison des politiques
87
+
88
+
En Reinforcement Learning
89
+
90
+
]
91
+
92
+
#let loop = (pauses: false) => diagram(
93
+
node((0, 0), $s_t$),
94
+
if pauses { pause } else { none },
95
+
if pauses { edge(corner: right, label-pos: 2 / 8, label-side: left)[choix de l'action] } else { none },
96
+
edge("->", corner: right, label-pos: 3 / 8, label-side: left)[$Pi$],
97
+
node((1, -1))[$a_t$],
98
+
if pauses { pause } else { none },
99
+
edge("->", corner: right, label-pos: 5 / 8, label-side: left)[$M$],
100
+
if pauses { edge(corner: right, label-pos: 6 / 8, label-side: left)[simulation] } else { none },
101
+
node((2, 0))[$s_(t+1)$],
102
+
if pauses { pause } else { none },
103
+
edge((2, 0), (2, .75), (0, .75), (0, 0), "-->", label-side: left)[itération],
104
+
)
105
+
106
+
#centered(loop(pauses: true))
107
+
#pagebreak()
108
+
109
+
110
+
#centered[
111
+
#grid(
112
+
columns: 2,
113
+
gutter: 3em,
114
+
115
+
loop(pauses: false),
116
+
117
+
[
118
+
#diagram(
119
+
$
120
+
s_0 edge(a_0, ->) & s_1 edge(a_1, ->) & s_2 edge(a_2, ->) & dots.c
121
+
$,
122
+
)
123
+
124
+
#pause
125
+
126
+
$
127
+
((s_0, a_0), (s_1, a_1), (s_2, a_2), ...)
128
+
pause
129
+
in cal(C)
130
+
$
131
+
132
+
],
133
+
)
134
+
]
135
+
136
+
#pagebreak()
137
+
138
+
#centered[
139
+
$
140
+
A & := "actions possibles" \
141
+
S & := "états possibles" \
142
+
#definitions_paths_set
143
+
$
144
+
]
145
+
146
+
#pagebreak()
147
+
148
+
== Comparaison des politiques: Avantage $A$
149
+
150
+
#centered[
151
+
À quel point est-il mieux de choisir $a_t$ plutôt qu'une autre action?
152
+
]
153
+
154
+
#pagebreak()
155
+
156
+
#centered[
157
+
158
+
159
+
#let height = 2
160
+
#scale(70%, reflow: true, diagram((
161
+
// Prior path
162
+
node((0, 0))[$dots.c$],
163
+
edge("->")[$a_(t-2)$],
164
+
node((1, 0))[$s_(t-1)$],
165
+
edge("->")[$a_(t-1)$],
166
+
pause,
167
+
node((2, 0), name: <break>)[$s_t$],
168
+
edge("-")[],
169
+
node((3.5, 0)),
170
+
edge("->", label-pos: 0%)[$a_t$],
171
+
node((4.5, 0))[$sum_(i=t+1)^oo gamma^t r(s_i)$],
172
+
node(name: <bottom>, (4.5, +1.5))[$sum_(i=t+1)^oo gamma^t r(s'_i)$ ],
173
+
node(name: <top>, (4.5, -1.5))[$sum_(i=t+1)^oo gamma^t r(s''_i)$],
174
+
edge(<break>, <bottom>, "->", bend: -25deg)[$a'_t$],
175
+
edge(<break>, <top>, "->", bend: 25deg)[$a''_t$],
176
+
pause,
177
+
// Expectation bar V(s)
178
+
node((5, height)),
179
+
edge("--"),
180
+
node((1.85, height)),
181
+
edge("-", label-side: left, label-pos: 75%)[$exp$],
182
+
node((1.85, -height)),
183
+
edge("--")[],
184
+
node((5, -height)),
185
+
// Expectation bar Q(s, a)
186
+
node((5, 0.5)),
187
+
edge("--"),
188
+
node((3.25, 0.5)),
189
+
edge("-", label-side: left, label-pos: 75%)[$exp$],
190
+
node((3.25, -0.5)),
191
+
edge("--")[],
192
+
node((5, -0.5)),
193
+
)))
194
+
195
+
#pause
196
+
197
+
$
198
+
A_(pi, r)(s, a) := exp("avec" a_t) - exp("à" thick t-1)
199
+
$
200
+
201
+
]
202
+
203
+
#pagebreak()
204
+
205
+
206
+
#what_is_surrogate_adv