Files
tp1-iaavancee/ex1.py
Namu 3d8dab7f06
All checks were successful
SonarQube Scan / SonarQube Trigger (push) Successful in 23s
First commit
2025-09-19 18:20:43 +02:00

88 lines
2.1 KiB
Python

import numpy as np
import pandas as pd
from typing import List
def best_path(start_label: str, goal_label: str) -> List[str]:
s = labels.index(start_label)
g = labels.index(goal_label)
path = [start_label]
while s!= g:
a = np.argmax(Q[s]) # ici, on récupère l'action
next = NEXT[s, a]
if next is None:
raise f'Action impossible State{s} Action{a} NextState{next}'
s = labels.index(next)
path.append(next)
return path
labels = list("ABCDEFGHIJKL")
R = np.array([
# UP DOWN LEFT RIGHT
[0,0,0,1], #A
[0,1,1,1], #B
[0,1,1,0], #C
[0,1,0,0], #D
[0,1,0,0], #E
[1,1,0,0], #F
[1,0,0,1], #G
[1,1,1,0], #H
[1,0,0,1], #I
[1,0,1,1], #J
[0,0,1,1], #K
[1,0,1,0], #L
], dtype=float)
# Fait une matrice de même dimension que R remplie de 0
Q = np.zeros_like(R)
# on a l'état courant et l'action en cours, il nous faut st+1 (la prochaine action)
NEXT = np.array([
# UP DOWN LEFT RIGHT
[None, None, None, 'B'], #A
[None, 'F', 'A', 'C'], #B
[None, 'G', 'B', None], #C
[None, 'H', None, None], #D
[None, 'I', None, None], #E
['B', 'J', None, None], #F
['C', None, None, 'H'], #G
['D', 'L', 'G', None], #H
['E', None, None, 'J'], #I
['F', None, 'I', 'K'], #J
[None, None, 'J', 'L'], #K
['H', None, 'K', None], #L
])
# Hyperparameters
gamma = 0.75
alpha = 0.90
n_iters = 1_000
rng = np.random.default_rng(0)
# augmente le reward pour les directions qui mènent à G (C DOWN & H LEFT)
goal_opt1 = labels.index('C')
down_index = 1
goal_opt2 = labels.index('H')
left_index = 2
R_goal = R.copy()
R_goal[goal_opt1, down_index] = 1_000.0
R_goal[goal_opt2, left_index] = 1_000.0
for _ in range(n_iters):
s = rng.integers(0, R.shape[0]) # random current state
actions = np.where(R_goal[s] > 0)[0] # valid actions
if actions.size == 0:
continue
a = rng.choice(actions) # random valid action
s_next = a # transition to next state
TD = R_goal[s, a] + gamma * Q[s_next].max() - Q[s, a]
Q[s, a] += alpha * TD
print("Matrice Q: ")
print(Q)
print("Path E -> G: ", " -> ".join(best_path('E', 'G')))