def test_fpccn_decompress():
"""Test the FPCNN class decompression function."""
data_shape = (8, 8, 8) # x, y, lambda
hyperparams = {
"track_spatial_length":
1,
"track_spatial_width":
5,
"track_spectral_length":
1,
"track_spectral_width":
5,
"track_fusion_length":
1,
"track_fusion_width":
5,
"lr":
0.01,
"context_offsets_spatial": [
(-1, 0, 0),
(-1, -1, 0),
(0, -1, 0),
(1, -1, 0),
(-1, 0, -1),
(-1, -1, -1),
(0, -1, -1),
(1, -1, -1),
(-1, 0, -2),
(-1, -1, -2),
(0, -1, -2),
(1, -1, -2),
],
"context_offsets_spectral": [(0, 0, -1), (0, 0, -2), (0, 0, -3),
(0, 0, -4)],
}
rng = np.random.default_rng()
data = rng.integers(low=0, high=16383, size=data_shape)
LOG.info(f"Data ({data.shape}, {data.dtype}):\n{data}")
model = models.FPCNN(hp=hyperparams, logname="test_decompress_toy")
output = model.decompress(data=data)
LOG.info(f"Output ({output.shape}, {output.dtype}):\n{output}")
def _build_model(self, params):
LOG.debug("Building model")
model = models.FPCNN(
hp={
"track_spatial_length":
1,
"track_spatial_width":
5,
"track_spectral_length":
1,
"track_spectral_width":
5,
"track_fusion_length":
1,
"track_fusion_width":
5,
"lr":
0.01,
"context_offsets_spatial": [
(-1, 0, 0),
(-1, -1, 0),
(0, -1, 0),
(1, -1, 0),
(-1, 0, -1),
(-1, -1, -1),
(0, -1, -1),
(1, -1, -1),
(-1, 0, -2),
(-1, -1, -2),
(0, -1, -2),
(1, -1, -2),
],
"context_offsets_spectral": [
(0, 0, -1),
(0, 0, -2),
(0, 0, -3),
(0, 0, -4),
],
},
logname="tuning_ip",
)
return model
def test_fpcnn_instantiate_model():
"""Test the FPCNN class model instantiation method."""
hyperparams = {
"track_spatial_length":
1,
"track_spatial_width":
5,
"track_spectral_length":
1,
"track_spectral_width":
5,
"track_fusion_length":
1,
"track_fusion_width":
5,
"lr":
0.01,
"context_offsets_spatial": [
(-1, 0, 0),
(-1, -1, 0),
(0, -1, 0),
(1, -1, 0),
(-1, 0, -1),
(-1, -1, -1),
(0, -1, -1),
(1, -1, -1),
(-1, 0, -2),
(-1, -1, -2),
(0, -1, -2),
(1, -1, -2),
],
"context_offsets_spectral": [(0, 0, -1), (0, 0, -2), (0, 0, -3),
(0, 0, -4)],
}
models.FPCNN(hp=hyperparams, logname="test_init")
(-1, 0, -2),
(-1, -1, -2),
(0, -1, -2),
(1, -1, -2),
],
"context_offsets_spectral": [(0, 0, -1), (0, 0, -2), (0, 0, -3), (0, 0, -4)],
}
# endregion
# load data
LOG.info(f"Loading data from '{data_path}'")
data = datalib.load_data_hdf5(path=data_path, header=data_header)
data = data[0 : data_shape[0], 0 : data_shape[1], 0 : data_shape[2]]
# instantiate compressor model
model = models.FPCNN(hp=hyperparams, logname="compress_ip")
# get initial weights and biases to encode later
weights = model.get_weights()
data_compressed = model.compress(data=data)
LOG.info(
f"Data compressed ({data_compressed.shape}, {data_compressed.dtype}):\n"
f"{data_compressed}"
)
# mapping
data_mapped = encoding.map_residuals(data_compressed)