File size: 6,562 Bytes
4484b8a |
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 |
import numpy as np
import torch
from torch.utils.data import DataLoader
from sklearn.model_selection import train_test_split
from torch.utils.data.dataset import IterableDataset
from collections import deque
from numpy.random import default_rng
DATA = np.load(
# "/home/seb/Perso/git/sudoku/sudoku_clean/data/sudoku_reshaped_million.npz"
"sudoku_reshaped_3_million.npz"
)
rng = np.random.default_rng()
def get_datasets(
add_proba_fill=False, train_size=1280 // 2, test_size=1280 // 2, max_holes=None
):
quizzes = DATA["quizzes"][: train_size + test_size]
solutions = DATA["solutions"][: train_size + test_size]
X = quizzes
if max_holes:
while True:
x_holes = X[:, 1].sum(-1) == 0
x_nb_holes = x_holes.sum((1, 2))
mask_x_max_holes = x_nb_holes > max_holes
if not any(mask_x_max_holes):
break
for idx_x in np.nonzero(mask_x_max_holes)[0]:
sub_x_holes = x_holes[idx_x]
idx_fill = rng.choice(np.transpose(np.nonzero(sub_x_holes)))
X[idx_x, :, idx_fill[0], idx_fill[1], :] = solutions[
idx_x, :, idx_fill[0], idx_fill[1], :
]
X = X.reshape(X.shape[0], 2, 9 * 9 * 9)
solutions = solutions.reshape(solutions.shape[0], 2, 9 * 9 * 9)
X_train, X_test, solutions_train, solutions_test = train_test_split(
X, solutions, test_size=test_size, random_state=42
)
if add_proba_fill:
X_train_bis = X_train.copy()
mask = solutions_train == 1
X_train_bis[mask] = np.random.randint(0, 2, size=mask.sum())
X_train = np.concatenate([X_train, X_train_bis])
solutions_train = np.concatenate([solutions_train, solutions_train])
train = torch.utils.data.TensorDataset(
torch.Tensor(X_train), torch.Tensor(solutions_train)
)
test = torch.utils.data.TensorDataset(
torch.Tensor(X_test), torch.Tensor(solutions_test)
)
return train, test
train_dataset, test_dataset = get_datasets()
def data_loader(batch_size=32, add_proba_fill=False):
train, test = get_datasets(add_proba_fill=add_proba_fill)
train_loader = torch.utils.data.DataLoader(train, batch_size=batch_size)
test_loader = torch.utils.data.DataLoader(test, batch_size=batch_size)
return train_loader, test_loader
class DataIterBuffer(IterableDataset):
def __init__(self, raw_dataset=[], buffer_optim=50, prop_new=0.1, seed=1):
self.raw_dataset = raw_dataset
# self.raw_dataset = iter(raw_dataset)
self.buffer = deque()
self.buffer_optim = buffer_optim
self.prop_new = prop_new
self.rng = default_rng(seed=seed)
self.idx_dataset = 0
def __iter__(self):
# while True:
# if (np.random.random() < self.prop_new) and (
# len(self.buffer) <= self.buffer_optim
# ):
# try:
# yield next(self.raw_dataset)
# except StopIteration:
# if len(self.buffer) != 0:
# yield self.buffer.popleft()
# else:
# break
# else:
# if len(self.buffer) != 0:
# yield self.buffer.popleft()
# else:
# try:
# yield next(self.raw_dataset)
# except StopIteration:
# break
while True:
if (np.random.random() < self.prop_new) and (
len(self.buffer) <= self.buffer_optim
):
if self.idx_dataset >= len(self.raw_dataset):
if len(self.buffer) != 0:
yield self.buffer.popleft()
else:
break
else:
yield self.raw_dataset[self.idx_dataset]
self.idx_dataset += 1
else:
if len(self.buffer) != 0:
yield self.buffer.popleft()
else:
if self.idx_dataset >= len(self.raw_dataset):
break
else:
yield self.raw_dataset[self.idx_dataset]
self.idx_dataset += 1
def append(self, X, Y) -> None:
"""Add experience to the buffer.
Args:
experience: tuple (state, action, reward, done, new_state)
"""
X[Y == 0] = 0
mask = ~(X == Y).view(-1, 2 * 729).all(dim=1)
for x, y in zip(X[mask], Y[mask]):
self.buffer.append((x, y))
def __len__(self):
return len(self.buffer) + len(self.raw_dataset)
# class DataIterDeepBuffer(IterableDataset):
# def __init__(self, raw_dataset=[], buffer_target_size=32, prop_new=0.1, seed=1, prof=6):
# self.raw_dataset = iter(raw_dataset)
# # self.buffer = deque()
# self.buffer_target_size = buffer_target_size
# self.prop_new = prop_new
# self.rng = default_rng(seed=seed)
# self.prof=prof
# self.buffers=[deque() for _ in range(prof)]
# def __iter__(self):
# while True:
# buffer_sizes = np.array([len(buffer) for buffer in self.buffers])
# if any(buffer_sizes>=self.buffer_target_size):
# #
# if (np.random.random() < self.prop_new) and (
# len(self.buffer) <= self.buffer_optim
# ):
# try:
# yield next(self.raw_dataset)
# except StopIteration:
# if len(self.buffer) != 0:
# yield self.buffer.popleft()
# else:
# break
# else:
# if len(self.buffer) != 0:
# yield self.buffer.popleft()
# else:
# try:
# yield next(self.raw_dataset)
# except StopIteration:
# break
# def append(self, X, Y) -> None:
# """Add experience to the buffer.
# Args:
# experience: tuple (state, action, reward, done, new_state)
# """
# X[Y == 0] = 0
# mask = ~(X == Y).view(-1, 2 * 729).all(dim=1)
# for x, y in zip(X[mask], Y[mask]):
# self.buffer.append((x, y))
# def __len__(self):
# return len(self.buffer) + len(self.raw_dataset)
|