|
import argparse |
|
import os |
|
from netCDF4 import Dataset |
|
|
|
|
|
def assemble_data(input_dir, output_file): |
|
nc_files = [f for f in os.listdir(input_dir) if f.endswith(".nc") and not f.startswith("c")] |
|
c_files = [f for f in os.listdir(input_dir) if f.endswith(".nc") and f.startswith("c")] |
|
nc_files.sort() |
|
c_files.sort() |
|
|
|
samples = [0] |
|
with Dataset(os.path.join(input_dir, nc_files[0]), "r") as first_nc: |
|
samples.append(first_nc.dimensions["sample"].size) |
|
num_times = first_nc.dimensions["time"].size |
|
try: |
|
num_channels = first_nc.dimensions["channel"].size |
|
except: |
|
num_channels = None |
|
x_size = first_nc.dimensions["x"].size |
|
y_size = first_nc.dimensions["y"].size |
|
dtype = first_nc.variables[nc_files[0].split("_")[0]].dtype |
|
|
|
for nc_file in nc_files[1:]: |
|
with Dataset(os.path.join(input_dir, nc_file), "r") as nc: |
|
samples.append(nc.dimensions["sample"].size) |
|
|
|
num_samples = sum(samples) |
|
for i in range(1, len(samples)): |
|
samples[i] += samples[i - 1] |
|
with Dataset(output_file, "w") as out_nc: |
|
out_nc.createDimension("sample", num_samples) |
|
out_nc.createDimension("time", num_times) |
|
if num_channels is not None: |
|
out_nc.createDimension("channel", num_channels) |
|
out_nc.createDimension("x", x_size) |
|
out_nc.createDimension("y", y_size) |
|
if num_channels is not None: |
|
out_nc.createVariable( |
|
nc_files[0].split("_")[0], dtype, ("sample", "time", "channel", "x", "y"), chunksizes=(1, 1, num_channels, x_size, y_size) |
|
) |
|
else: |
|
out_nc.createVariable( |
|
nc_files[0].split("_")[0], dtype, ("sample", "time", "x", "y"), chunksizes=(1, 1, x_size, y_size) |
|
) |
|
|
|
out_nc.createVariable( |
|
"c", dtype, ("sample", "x", "y"), chunksizes=(1, x_size, y_size) |
|
) |
|
|
|
for i, nc_file in enumerate(nc_files): |
|
with Dataset(os.path.join(input_dir, nc_file), "r") as nc: |
|
print(f"Processing {os.path.join(input_dir, nc_file)}") |
|
variable = nc.variables[nc_file.split("_")[0]] |
|
out_nc[nc_file.split("_")[0]][samples[i] : samples[i + 1]] = variable[:] |
|
|
|
samples = [0] |
|
for nc_file in c_files: |
|
with Dataset(os.path.join(input_dir, nc_file), "r") as nc: |
|
samples.append(nc.dimensions["sample"].size) |
|
for i in range(1, len(samples)): |
|
samples[i] += samples[i - 1] |
|
|
|
for i, c_file in enumerate(c_files): |
|
with Dataset(os.path.join(input_dir, c_file), "r") as nc: |
|
print(f"Processing {os.path.join(input_dir, c_file)}") |
|
variable = nc.variables[c_file.split("_")[0]] |
|
out_nc[c_file.split("_")[0]][samples[i] : samples[i + 1]] = variable[:] |
|
|
|
print(f"Saved data to {output_file}") |
|
|
|
|
|
if __name__ == "__main__": |
|
parser = argparse.ArgumentParser() |
|
parser.add_argument("--input_dir", type=str, required=True) |
|
parser.add_argument("--output_file", type=str, required=True) |
|
args = parser.parse_args() |
|
|
|
assemble_data(args.input_dir, args.output_file) |
|
|