def test_run_1dmackey():
bh.random.seed(0)
params = torsk.default_params()
params.input_map_specs = [{
"type": "random_weights",
"size": [1000],
"input_scale": 1.
}]
params.spectral_radius = 1.5
params.density = 0.05
params.input_shape = [1, 1]
params.train_length = 2200
params.pred_length = 400
params.transient_length = 200
params.dtype = "float64"
params.reservoir_representation = "dense"
params.backend = "numpy"
params.train_method = "pinv_svd"
params.tikhonov_beta = 2.0
params.debug = False
params.imed_loss = False
logger = logging.getLogger(__name__)
level = "DEBUG" if params.debug else "INFO"
logging.basicConfig(level=level)
logging.getLogger("matplotlib").setLevel("INFO")
model = ESN(params)
mackey = mackey_sequence(N=3700)
mackey = normalize(mackey) * 2 - 1
mackey = mackey[:, bh.newaxis, bh.newaxis]
dataset = ImageDataset(mackey, params, scale_images=False)
model, outputs, pred_labels = torsk.train_predict_esn(model, dataset)
# import matplotlib.pyplot as plt
# plt.plot(bh.squeeze(outputs))
# plt.plot(bh.squeeze(pred_labels))
# plt.show()
error = bh.abs(outputs - pred_labels)
logger.info(error.mean())
logger.info(error.max())
assert error.mean() < 0.2
assert error.max() < 1.1
params.update(sys.argv[1:])
logger.info(params)
if params.backend == "numpy":
logger.info("Running with NUMPY backend")
from torsk.data.numpy_dataset import NumpyImageDataset as ImageDataset
from torsk.models.numpy_esn import NumpyESN as ESN
else:
logger.info("Running with TORCH backend")
from torsk.data.torch_dataset import TorchImageDataset as ImageDataset
from torsk.models.torch_esn import TorchESN as ESN
npypath = pathlib.Path(
"/home/niklas/erda_save/Ocean/esn/Kuro_SSH_5daymean.npy")
images = np.load(npypath)[:, 90:190, 90:190]
images[images > 10000.] = 0.
images = resample2d_sequence(images, params.input_shape)
dataset = ImageDataset(images, params, scale_images=True)
logger.info("Building model ...")
model = ESN(params)
logger.info("Training + predicting ...")
model, outputs, pred_labels = torsk.train_predict_esn(
model,
dataset,
"/home/niklas/erda_save/kuro_conv_5daymean",
steps=1000,
step_length=1)
from torsk.data.numpy_dataset import NumpyImageDataset as ImageDataset
from torsk.models.numpy_esn import NumpyESN as ESN
else:
logger.info("Running with TORCH backend")
from torsk.data.torch_dataset import TorchImageDataset as ImageDataset
from torsk.models.torch_esn import TorchESN as ESN
logger.info(params)
logger.info("Creating circle dataset ...")
t = np.arange(0, 200 * np.pi, 0.02 * np.pi)
x, y = np.sin(0.3 * t), np.cos(t)
center = np.array([y, x]).T
images = gauss2d_sequence(center, sigma=0.5, size=params.input_shape)
dataset = ImageDataset(images, params)
logger.info("Building model ...")
model = ESN(params)
logger.info("Training + predicting ...")
model, outputs, pred_labels = torsk.train_predict_esn(
model, dataset, outdir="rand_output", steps=1)
logger.info("Visualizing results ...")
if params.backend == "torch":
pred_labels = pred_labels.squeeze().numpy()
outputs = outputs.squeeze().numpy()
anim = animate_double_imshow(pred_labels, outputs)
plt.show()
def test_lissajous(tmpdir):
np.random.seed(0)
params = torsk.default_params()
params.input_map_specs = [{
"type": "pixels",
"size": [30, 30],
"input_scale": 3.
}, {
"type": "conv",
"mode": "same",
"size": [5, 5],
"kernel_type": "gauss",
"input_scale": 4.
}, {
"type": "conv",
"mode": "same",
"size": [10, 10],
"kernel_type": "gauss",
"input_scale": 3.
}, {
"type": "conv",
"mode": "same",
"size": [15, 15],
"kernel_type": "gauss",
"input_scale": 4.
}, {
"type": "conv",
"mode": "same",
"size": [5, 5],
"kernel_type": "random",
"input_scale": 4.
}, {
"type": "conv",
"mode": "same",
"size": [10, 10],
"kernel_type": "random",
"input_scale": 4.
}, {
"type": "conv",
"mode": "same",
"size": [20, 20],
"kernel_type": "random",
"input_scale": 4.
}, {
"type": "dct",
"size": [15, 15],
"input_scale": 1.
}, {
"type": "gradient",
"input_scale": 4.
}, {
"type": "gradient",
"input_scale": 4.
}]
params.spectral_radius = 2.0
params.density = 0.001
params.input_shape = [30, 30]
params.train_length = 1200
params.pred_length = 300
params.transient_length = 200
params.dtype = "float64"
params.reservoir_representation = "sparse"
params.backend = "numpy"
params.train_method = "pinv_svd"
params.imed_loss = False
params.tikhonov_beta = None
params.debug = False
logger = logging.getLogger(__name__)
level = "DEBUG" if params.debug else "INFO"
logging.basicConfig(level=level)
logging.getLogger("matplotlib").setLevel("INFO")
logger.info("Creating circle dataset ...")
t = np.arange(0, 200 * np.pi, 0.02 * np.pi)
x, y = np.sin(0.3 * t), np.cos(t)
center = np.array([y, x]).T
images = gauss2d_sequence(center, sigma=0.5, size=params.input_shape)
dataset = ImageDataset(images, params, scale_images=True)
logger.info("Building model ...")
model = ESN(params)
logger.info("Training + predicting ...")
model, outputs, pred_labels = torsk.train_predict_esn(model, dataset)
# logger.info("Visualizing results ...")
# import matplotlib.pyplot as plt
# from torsk.visualize import animate_double_imshow
# anim = animate_double_imshow(pred_labels, outputs)
# plt.show()
error = np.abs(outputs - pred_labels)
logger.info(error.mean())
logger.info(error.max())
assert error.mean() < 3e-4
assert error.max() < 9e-3
if params.backend == "numpy":
logger.info("Running with NUMPY backend")
from torsk.data.numpy_dataset import NumpyImageDataset as ImageDataset
from torsk.models.numpy_esn import NumpyESN as ESN
else:
logger.info("Running with TORCH backend")
from torsk.data.torch_dataset import TorchImageDataset as ImageDataset
from torsk.models.torch_esn import TorchESN as ESN
logger.info("Building model ...")
model = ESN(params)
mackey, _ = mackey_anomaly_sequence(
N=3700,
anomaly_start=params.anomaly_start,
anomaly_step=params.anomaly_step)
mackey = normalize(mackey) * 2 - 1
mackey = mackey[:, np.newaxis, np.newaxis]
dataset = ImageDataset(mackey, params, scale_images=False)
logger.info("Training + predicting ...")
model, outputs, pred_labels = torsk.train_predict_esn(
model, dataset, "mackey_anomaly_output",
steps=1, step_length=2)
plt.plot(pred_labels[:, 0, 0], label="Prediction")
plt.plot(outputs[:, 0, 0], label="Truth")
plt.legend()
plt.show()
def test_kuro():
np.random.seed(0)
params = torsk.default_params()
params.input_map_specs = [{
"type": "pixels",
"size": [30, 30],
"input_scale": 3.
}, {
"type": "conv",
"mode": "same",
"size": [5, 5],
"kernel_type": "gauss",
"input_scale": 2.
}, {
"type": "conv",
"mode": "same",
"size": [10, 10],
"kernel_type": "gauss",
"input_scale": 1.5
}, {
"type": "conv",
"mode": "same",
"size": [15, 15],
"kernel_type": "gauss",
"input_scale": 1.
}, {
"type": "conv",
"mode": "same",
"size": [5, 5],
"kernel_type": "random",
"input_scale": 1.
}, {
"type": "conv",
"mode": "same",
"size": [10, 10],
"kernel_type": "random",
"input_scale": 1.
}, {
"type": "conv",
"mode": "same",
"size": [20, 20],
"kernel_type": "random",
"input_scale": 1.
}, {
"type": "dct",
"size": [15, 15],
"input_scale": 1.
}, {
"type": "gradient",
"input_scale": 1.
}, {
"type": "gradient",
"input_scale": 1.
}]
params.spectral_radius = 1.5
params.density = 0.01
params.input_shape = [30, 30]
params.train_length = 12 * 73
params.pred_length = 73
params.transient_length = 3 * 73
params.dtype = "float64"
params.reservoir_representation = "sparse"
params.backend = "numpy"
params.train_method = "pinv_lstsq"
params.tikhonov_beta = 3e1
params.debug = False
params.imed_loss = True
params.imed_sigma = 1.0
logger = logging.getLogger(__name__)
level = "DEBUG" if params.debug else "INFO"
logging.basicConfig(level=level)
logging.getLogger("matplotlib").setLevel("INFO")
images = np.load(pathlib.Path(__file__).parent / "kuro_test_sequence.npy")
dataset = ImageDataset(images, params, scale_images=True)
logger.info("Building model ...")
model = ESN(params)
logger.info("Training + predicting ...")
model, outputs, pred_labels = torsk.train_predict_esn(model, dataset)
# logger.info("Visualizing results ...")
# import matplotlib.pyplot as plt
# from torsk.visualize import animate_double_imshow
# anim = animate_double_imshow(pred_labels, outputs)
# plt.show()
error = np.abs(outputs - pred_labels)
logger.info(error.mean())
logger.info(error.max())
assert error.mean() < 0.2
assert error.max() < 1.3