Skip to content

GitLab

Commit 1df8a46a authored by tpys's avatar tpys
Browse files

clean and rename

parent b5342442
Show whitespace changes
Inline Side-by-side
Showing with 239 additions and 12 deletions
README.md
View file @ 1df8a46a
...@@ -21,10 +21,12 @@ The downloaded files shall be organized as the following hierarchy: ...@@ -21,10 +21,12 @@ The downloaded files shall be organized as the following hierarchy:
│ ├── data │ ├── data
│ │ ├── 20210101 │ │ ├── 20210101
│ │ ├── input.nc │ │ ├── input.nc
│ │ ├── output.nc │ │ ├── output
│ │ ├── target.nc │ │ ├── 006.nc
│ │ │ │ ├── 012.nc
│ ├── fuxi │ │ ├── ...
│ │ ├── 360.nc
│ ├── model
│ | ├── short │ | ├── short
│ | ├── short.onnx │ | ├── short.onnx
│ | ├── medium │ | ├── medium
...@@ -32,7 +34,11 @@ The downloaded files shall be organized as the following hierarchy: ...@@ -32,7 +34,11 @@ The downloaded files shall be organized as the following hierarchy:
│ | ├── long │ | ├── long
│ | ├── long.onnx │ | ├── long.onnx
| | | |
│ ├── infernece_fuxi.py │ ├── fuxi.py
│ ├── util.py
│ ├── make_era5_input.py
│ ├── make_hres_input.py
│ ├── make_gfs_input.py
``` ```
...@@ -51,7 +57,7 @@ pip install -r requirement.txt ...@@ -51,7 +57,7 @@ pip install -r requirement.txt
## Demo ## Demo
```bash ```bash
python inference_fuxi.py --model model_dir --input input_file --num_steps 20 --input_type GFS python fuxi.py --model model_dir --input input_file --num_steps 20 20 20
``` ```
......
inference_fuxi.py → fuxi.py
View file @ 1df8a46a
...@@ -6,22 +6,18 @@ import xarray as xr ...@@ -6,22 +6,18 @@ import xarray as xr
import pandas as pd import pandas as pd
import onnxruntime as ort import onnxruntime as ort
from data_util import save_like from util import save_like
ort.set_default_logger_severity(3) ort.set_default_logger_severity(3)
parser = argparse.ArgumentParser() parser = argparse.ArgumentParser()
parser.add_argument('--model', type=str, required=True, help="FuXi onnx model dir") parser.add_argument('--model', type=str, required=True, help="FuXi onnx model dir")
parser.add_argument('--input', type=str, required=True, help="The input data file, store in netcdf format") parser.add_argument('--input', type=str, required=True, help="The input data file, store in netcdf format")
parser.add_argument('--input_type', type=str, help="The input type", default="ERA5")
parser.add_argument('--save_dir', type=str, default="") parser.add_argument('--save_dir', type=str, default="")
parser.add_argument('--num_steps', type=int, nargs="+", default=[20]) parser.add_argument('--num_steps', type=int, nargs="+", default=[20])
args = parser.parse_args() args = parser.parse_args()
assert args.input_type.upper() in ["ERA5", "GFS", "HRES"]
def time_encoding(init_time, total_step, freq=6): def time_encoding(init_time, total_step, freq=6):
init_time = np.array([init_time]) init_time = np.array([init_time])
tembs = [] tembs = []
...@@ -91,7 +87,7 @@ def run_inference(model_dir, data, num_steps, save_dir=""): ...@@ -91,7 +87,7 @@ def run_inference(model_dir, data, num_steps, save_dir=""):
temb = tembs[step] temb = tembs[step]
new_input, = session.run(None, {'input': input, 'temb': temb}) new_input, = session.run(None, {'input': input, 'temb': temb})
output = new_input[:, -1] output = new_input[:, -1]
save_like(output, data, step, save_dir, input_type=args.input_type) save_like(output, data, step, save_dir)
print(f'stage: {i}, step: {step+1:02d}, output: {output.min():.2f} {output.max():.2f}') print(f'stage: {i}, step: {step+1:02d}, output: {output.min():.2f} {output.max():.2f}')
input = new_input input = new_input
step += 1 step += 1
......
make_hres_input.py 0 → 100644
View file @ 1df8a46a
import os
import numpy as np
import pandas as pd
import xarray as xr
def make_hres_input(init_time, data_dir, save_dir, degree=0.25):
lat = np.linspace(-90, 90, int(180/degree)+1, dtype=np.float32)
lon = np.arange(0, 360, degree, dtype=np.float32)
pl_names = ['z', 't', 'u', 'v', 'r']
sfc_names = ['t2m', 'u10', 'v10', 'msl', 'tp']
levels = [50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000]
input = []
level = []
for name in pl_names + sfc_names:
src_name = '{}_{}'.format(name, init_time.strftime("%Y%m%d%H.nc"))
src_file = os.path.join(data_dir, src_name)
if not os.path.exists(src_file):
return
try:
v = xr.open_dataset(src_file)
v = v.sel(time=init_time, drop=True).data
except:
print(f"open {src_file} failed")
return
# is there nan in raw data ?
if np.isnan(v).sum() > 0:
print(f"{src_name} has nan value")
return
# interpolate to 0.25 deg
v = v.interp(lat=lat, lon=lon, kwargs={"fill_value": "extrapolate"})
# make sure on nan
if np.isnan(v).sum() > 0:
print(f"{src_name} has nan value")
return
# reverse pressure level
try:
if name in pl_names:
v = xr.concat([v.sel(level=l) for l in levels], 'level')
level.extend([f'{name}{l}' for l in levels])
except:
print("missing pressure level")
return
if name in sfc_names:
level.append(name)
# temperature in kelvin
if name == "t":
v = v + 273.15
# FuXi take two step as input
if name == "tp":
v = v.clip(min=0, max=1000)
zero = v * 0
zero = zero.assign_coords(dtime=[0])
v = xr.concat([zero, v], "dtime")
print(f'{src_name}: {v.min().values:.2f} ~ {v.max().values:.2f}')
v.attrs = {}
v = v.rename({'dtime': 'time'})
v = v.squeeze('member').drop('member')
input.append(v)
# concat and reshape
input = xr.concat(input, "level")
input = input.transpose("time", "level", "lat", "lon")
valid_time = init_time + pd.Timedelta(hours=6) # utc time
v = v.assign_coords(time=[init_time, valid_time])
# reverse latitude
input = input.reindex(lat=input.lat[::-1])
input = input.assign_coords(level=level)
input.name = 'data'
# save to nc
print(input)
save_name = os.path.join(save_dir, init_time.strftime("%Y%m%d-%H.nc"))
input = input.astype(np.float32)
input.to_netcdf(save_name)
def test_make_input():
init_time = pd.to_datetime("20230731-12") # must utc
data_dir = "data/HRES"
save_dir = "data/HRES/input"
os.makedirs(save_dir, exist_ok=True)
make_hres_input(init_time, data_dir, save_dir)
data_util.py → util.py
View file @ 1df8a46a
...@@ -9,7 +9,6 @@ __all__ = ["save_like"] ...@@ -9,7 +9,6 @@ __all__ = ["save_like"]
pl_names = ['z', 't', 'u', 'v', 'r'] pl_names = ['z', 't', 'u', 'v', 'r']
sfc_names = ['t2m', 'u10', 'v10', 'msl', 'tp'] sfc_names = ['t2m', 'u10', 'v10', 'msl', 'tp']
levels = [50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000] levels = [50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000]
degree = 0.25
def weighted_rmse(out, tgt): def weighted_rmse(out, tgt):
...@@ -33,16 +32,12 @@ def split_variable(ds, name): ...@@ -33,16 +32,12 @@ def split_variable(ds, name):
return v return v
def save_like(output, input, step, save_dir="", input_type="hres", freq=6, split=False): def save_like(output, input, step, save_dir="", freq=6, split=False):
if save_dir: if save_dir:
os.makedirs(save_dir, exist_ok=True) os.makedirs(save_dir, exist_ok=True)
step = (step+1) * freq step = (step+1) * freq
init_time = pd.to_datetime(input.time.values[-1]) init_time = pd.to_datetime(input.time.values[-1])
if input_type.upper() == "HRES":
step = (step+2) * freq
init_time = pd.to_datetime(input.time.values[0])
ds = xr.DataArray( ds = xr.DataArray(
output[None], output[None],
dims=['time', 'step', 'level', 'lat', 'lon'], dims=['time', 'step', 'level', 'lat', 'lon'],
...@@ -75,252 +70,6 @@ def save_like(output, input, step, save_dir="", input_type="hres", freq=6, split ...@@ -75,252 +70,6 @@ def save_like(output, input, step, save_dir="", input_type="hres", freq=6, split
ds.to_netcdf(save_name) ds.to_netcdf(save_name)
def make_era5_input(init_time, data_dir, save_dir):
ds = []
init_time = pd.to_datetime(init_time)
hist_time = init_time - pd.Timedelta(hours=6)
print(f"init_time: {init_time}")
level = []
for name in pl_names + sfc_names:
data_name = os.path.join(data_dir, name, f'{init_time.year}')
v = xr.open_zarr(data_name)
v = v.sel(time=[hist_time, init_time])
v = v.rename({name: 'data'})
v.attrs = {}
ds.append(v)
if name in pl_names:
level.extend([f'{name.lower()}{l}' for l in levels])
if name in sfc_names:
level.append(name.lower())
ds = xr.concat(ds, 'level')
ds = ds.assign_coords(level=level)
os.makedirs(save_dir, exist_ok=True)
save_name = os.path.join(save_dir, init_time.strftime("input.%Y%m%d.t%H.nc"))
print(f"save to {save_name} ...")
ds = ds.astype(np.float32)
ds.to_netcdf(save_name)
def make_era5(init_time, data_dir):
import os
import numpy as np
import pandas as pd
import xarray as xr
init_time = pd.to_datetime(init_time)
print(f"process {init_time} ...")
pl_file = os.path.join(data_dir, init_time.strftime('P%Y%m%d%H.nc'))
pl = xr.open_dataset(pl_file)
sfc_file = os.path.join(data_dir, init_time.strftime('S%Y%m%d%H.nc'))
sfc = xr.open_dataset(sfc_file)
tp_file = os.path.join(data_dir, init_time.strftime('R%Y%m%d.nc'))
tp = xr.open_dataarray(tp_file).fillna(0)
tp = tp.rolling(time=6).sum() * 1000
tp = tp.sel(time=tp.time[::6])
tp = tp.clip(min=0, max=1000)
sfc['tp'] = tp
pl_names = ['z', 't', 'u', 'v', 'r']
sfc_names = ['t2m', 'u10', 'v10', 'msl', 'tp']
levels = [50, 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, 1000]
channel = [f'{n.upper()}{l}' for n in pl_names for l in levels]
channel +=[n.upper() for n in sfc_names]
ds = []
for name in pl_names + sfc_names:
if name in ['z', 't', 'u', 'v', 'r']:
v = pl[name]
if name in ['t2m', 'u10', 'v10', 'msl', 'tp']:
v = sfc[name]
level = xr.DataArray([1], coords={'level': [1]}, dims=['level'])
v = v.expand_dims({'level': level}, axis=1)
if np.isnan(v).sum() > 0:
print(f"{name} has nan value")
raise ValueError
v.name = "data"
v.attrs = {}
print(f"{name}: {v.shape}, {v.min().values} ~ {v.max().values}")
ds.append(v)
ds = xr.concat(ds, 'level')
ds = ds.assign_coords(level=channel)
ds = ds.rename({'longitude': 'lon', 'latitude': 'lat'})
ds = ds.astype(np.float32)
return ds
# ds12 = make_era5('20230725-12', 'ERA520230725')
# ds18 = make_era5('20230725-18', 'ERA520230725')
# ds = xr.concat([ds12, ds18], 'time')
# ds.to_netcdf('new_input.nc')
def make_hres_input(init_time, data_dir, save_dir):
lat = np.linspace(-90, 90, int(180/degree)+1, dtype=np.float32)
lon = np.arange(0, 360, degree, dtype=np.float32)
input = []
level = []
for name in pl_names + sfc_names:
src_name = '{}_{}'.format(name, init_time.strftime("%Y%m%d%H.nc"))
src_file = os.path.join(data_dir, src_name)
if not os.path.exists(src_file):
return
try:
v = xr.open_dataset(src_file)
v = v.sel(time=init_time, drop=True).data
except:
print(f"open {src_file} failed")
return
# is there nan in raw data ?
if np.isnan(v).sum() > 0:
print(f"{src_name} has nan value")
return
# interpolate to 0.25 deg
v = v.interp(lat=lat, lon=lon, kwargs={"fill_value": "extrapolate"})
# make sure on nan
if np.isnan(v).sum() > 0:
print(f"{src_name} has nan value")
return
# reverse pressure level
try:
if name in pl_names:
v = xr.concat([v.sel(level=l) for l in levels], 'level')
level.extend([f'{name}{l}' for l in levels])
except:
print("missing pressure level")
return
if name in sfc_names:
level.append(name)
# temperature in kelvin
if name == "t":
v = v + 273.15
# FuXi take two step as input
if name == "tp":
v = v.clip(min=0, max=1000)
zero = v * 0
zero = zero.assign_coords(dtime=[0])
v = xr.concat([zero, v], "dtime")
print(f'{src_name}: {v.min().values:.2f} ~ {v.max().values:.2f}')
v.attrs = {}
v = v.rename({'dtime': 'time'})
v = v.squeeze('member').drop('member')
input.append(v)
# concat and reshape
input = xr.concat(input, "level")
input = input.transpose("time", "level", "lat", "lon")
valid_time = init_time + pd.Timedelta(hours=6) # utc time
v = v.assign_coords(time=[init_time, valid_time])
# reverse latitude
input = input.reindex(lat=input.lat[::-1])
input = input.assign_coords(level=level)
input.name = 'data'
# save to nc
print(input)
save_name = os.path.join(save_dir, init_time.strftime("%Y%m%d-%H.nc"))
input = input.astype(np.float32)
input.to_netcdf(save_name)
def make_gfs_input(init_time, data_dir, save_dir):
pl_names = ['Z', 'T', 'U', 'V', 'R']
sfc_names = ['t2m', 'u10', 'v10', 'msl', 'tp']
lon = np.arange(0, 360, degree, dtype=np.float32)
lat = np.arange(90, -90, -degree, dtype=np.float32)
input = []
level = []
for name in pl_names + sfc_names:
src_name = '{}_{}'.format(name, init_time.strftime("%Y%m%d.nc"))
src_file = os.path.join(data_dir, src_name)
if not os.path.exists(src_file):
print(src_file)
return
try:
v = xr.open_dataset(src_file)[name]
except:
print(f"open {src_file} failed")
return
if np.isnan(v).sum() > 0:
print(f"{src_name} has nan value")
return
if v.shape[-2:] != (721, 1440):
v = v.interp(lat=lat, lon=lon, kwargs={
"fill_value": "extrapolate"})
if np.isnan(v).sum() > 0:
print(f"{src_name} has nan value")
return
if name in pl_names:
level.extend([f'{name.lower()}{l}' for l in levels])
if name in sfc_names:
level.append(name.lower())
if name == "Z":
v = v * 9.8
if name == "tp":
v = v.clip(min=0, max=1000)
v = v.squeeze('step').drop('step')
v.attrs = {}
v.name = 'data'
vmin = v.min().values
vmax = v.max().values
if vmax > 1e10:
v = v.where(v < 1e10, 0)
vmax = v.max().values
assert vmax < 1e10
print(f'{src_name}: {v.shape}, {vmin:.2f} ~ {vmax:.2f}')
input.append(v)
input = xr.concat(input, "level") # T
input = input.rename({"latitude": "lat", "longitude": "lon"})
times = [pd.to_datetime(str(t), format='%Y%m%d%H')
for t in input.time.values]
input = input.assign_coords(level=level)
input = input.assign_coords(time=times)
# TODO, we only need two time step input with dims: 2 x 70 x 721 x 1440
save_name = os.path.join(save_dir, init_time.strftime("%Y%m%d.nc"))
input = input.astype(np.float32)
input.to_netcdf(save_name)
def visualize(save_name, vars=[], titles=[], vmin=None, vmax=None): def visualize(save_name, vars=[], titles=[], vmin=None, vmax=None):
import cartopy.crs as ccrs import cartopy.crs as ccrs
import matplotlib.pyplot as plt import matplotlib.pyplot as plt
...@@ -354,14 +103,6 @@ def visualize(save_name, vars=[], titles=[], vmin=None, vmax=None): ...@@ -354,14 +103,6 @@ def visualize(save_name, vars=[], titles=[], vmin=None, vmax=None):
plt.close() plt.close()
def test_make_input():
init_time = pd.to_datetime("20230731-12") # must utc
data_dir = "data/HRES"
save_dir = "data/HRES/input"
os.makedirs(save_dir, exist_ok=True)
make_hres_input(init_time, data_dir, save_dir)
def test_visualize(step, data_dir): def test_visualize(step, data_dir):
src_name = os.path.join(data_dir, f"{step:03d}.nc") src_name = os.path.join(data_dir, f"{step:03d}.nc")
ds = xr.open_dataarray(src_name).isel(time=0) ds = xr.open_dataarray(src_name).isel(time=0)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment