import rrn_utils
train_set = Dataset.load('./data/without_sol_one.pkl')
test_set = Dataset.load('./data/with_sol_one.pkl')
hyperparameters = {
'device': 6,
'dim_x': 2,
'dim_y': 2,
'num_iters': 32,
'train_size': 3000,
'valid_size': 500,
'batch_size': 500,
'epochs': 600,
'valid_epochs': 25,
'save_epochs': 25,
'embed_size': 6,
'hidden_layer_size': 32,
'learning_rate': 1e-3,
'weight_decay': 1e-4
}
rrn_utils.train_rrn(hyperparameters,
train_inputs = train_set.get_input_data(0, 3000),
train_outputs = train_set.get_output_data(0, 3000),
other_inputs = { 'validation': train_set.get_input_data(3000, 3500),
'test': test_set.get_input_data(0, 500)},
other_outputs = { 'validation': train_set.get_output_data(3000, 3500),
'test': test_set.get_output_data(0, 500)})
'device': 4,
'dim_x': 2,
'dim_y': 2,
'num_iters': 32,
'train_size': train_size_per_num_hints * 12,
'valid_size': valid_size_per_num_hints * 12,
'batch_size': 500,
'epochs': 600,
'save_epochs': 25,
'embed_size': 6,
'hidden_layer_size': 32,
'learning_rate': 1e-3,
'weight_decay': 1e-4
}
dataset = Datasets.load('./data/datasets.pkl')
split_inputs, split_outputs = dataset.split_data([
train_size_per_num_hints,
train_size_per_num_hints + valid_size_per_num_hints
])
data = {
'train_inputs': split_inputs[0],
'train_outputs': split_outputs[0],
'valid_inputs': split_inputs[1],
'valid_outputs': split_outputs[1]
}
train_rrn(hyperparameters, data)
sys.path.append(SUDOKU_PATH + '/src/models')
from dataset import Dataset
from rrn_utils import train_rrn
train_set = Dataset.load('./data/train_set.pkl')
valid_set = Dataset.load('./data/valid_set.pkl')
devices = [4, 5, 6, 7, 8, 9]
print("Running on devices {}".format(devices))
hyperparameters = {
'devices': devices,
'dim_x': 3,
'dim_y': 3,
'num_iters': 32,
'batch_size': 32 * len(devices),
'epochs': 3,
'valid_epochs': 25,
'save_epochs': 25,
'embed_size': 16,
'hidden_layer_size': 96,
'learning_rate': 1e-4,
'weight_decay': 1e-4
}
train_rrn(hyperparameters,
train_inputs=train_set.keys()[:10000],
train_outputs=train_set.values()[:10000],
other_inputs={"validation": valid_set.keys()},
other_outputs={"validation": valid_set.values()})
hyperparameters = {
'device': 6,
'dim_x': 2,
'dim_y': 2,
'num_iters': 32,
'train_size': train_size_per_num_hints * 12,
'valid_size': valid_size_per_num_hints * 12,
'batch_size': 500,
'epochs': 600,
'valid_epochs': 25,
'save_epochs': 25,
'embed_size': 2,
'hidden_layer_size': 32,
'learning_rate': 1e-3,
'weight_decay': 1e-4
}
dataset = Datasets.load('../../4x4_all_reimbed/data/datasets.pkl')
split_inputs, split_outputs = dataset.split_data([
train_size_per_num_hints,
train_size_per_num_hints + valid_size_per_num_hints
])
rrn_utils.train_rrn(hyperparameters,
train_inputs=split_inputs[0],
train_outputs=split_outputs[0],
other_inputs={'validation': split_inputs[1]},
other_outputs={'validation': split_outputs[1]})
'batch_size': 400,
'epochs': 600,
'save_epochs': 25,
'embed_size': 6,
'hidden_layer_size': 32,
'learning_rate': 1e-3,
'weight_decay': 1e-4
}
dataset = Datasets.load('./data/train_datasets.pkl')
ext_valid_datasets = Datasets.load('./data/ext_valid_datasets.pkl')
split_inputs, split_outputs = dataset.split_data([
train_size_per_num_hints,
train_size_per_num_hints + valid_size_per_num_hints
])
extrapolation_inputs, extrapolation_outputs = ext_valid_datasets.split_data(
[valid_size_per_num_hints])
train_rrn(hyperparameters,
train_inputs=split_inputs[0],
train_outputs=split_outputs[0],
other_inputs={
"interpolation": split_inputs[1],
"extrapolation": extrapolation_inputs[0]
},
other_outputs={
"interpolation": split_outputs[1],
"extrapolation": extrapolation_outputs[0]
})