ocf-data-sampler 0.0.23__py3-none-any.whl → 0.0.25__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of ocf-data-sampler might be problematic. Click here for more details.
- ocf_data_sampler/config/model.py +34 -0
- ocf_data_sampler/load/load_dataset.py +55 -0
- ocf_data_sampler/load/nwp/providers/ecmwf.py +5 -2
- ocf_data_sampler/load/site.py +30 -0
- ocf_data_sampler/numpy_batch/__init__.py +4 -3
- ocf_data_sampler/numpy_batch/gsp.py +12 -12
- ocf_data_sampler/numpy_batch/nwp.py +14 -14
- ocf_data_sampler/numpy_batch/satellite.py +8 -8
- ocf_data_sampler/numpy_batch/site.py +29 -0
- ocf_data_sampler/select/__init__.py +8 -1
- ocf_data_sampler/select/dropout.py +2 -1
- ocf_data_sampler/select/geospatial.py +43 -1
- ocf_data_sampler/select/select_spatial_slice.py +8 -2
- ocf_data_sampler/select/spatial_slice_for_dataset.py +53 -0
- ocf_data_sampler/select/time_slice_for_dataset.py +124 -0
- ocf_data_sampler/time_functions.py +11 -0
- ocf_data_sampler/torch_datasets/process_and_combine.py +153 -0
- ocf_data_sampler/torch_datasets/pvnet_uk_regional.py +8 -418
- ocf_data_sampler/torch_datasets/site.py +196 -0
- ocf_data_sampler/torch_datasets/valid_time_periods.py +108 -0
- {ocf_data_sampler-0.0.23.dist-info → ocf_data_sampler-0.0.25.dist-info}/METADATA +1 -1
- ocf_data_sampler-0.0.25.dist-info/RECORD +66 -0
- {ocf_data_sampler-0.0.23.dist-info → ocf_data_sampler-0.0.25.dist-info}/WHEEL +1 -1
- {ocf_data_sampler-0.0.23.dist-info → ocf_data_sampler-0.0.25.dist-info}/top_level.txt +1 -0
- scripts/refactor_site.py +50 -0
- tests/conftest.py +62 -0
- tests/load/test_load_sites.py +14 -0
- tests/numpy_batch/test_gsp.py +1 -2
- tests/numpy_batch/test_nwp.py +1 -3
- tests/numpy_batch/test_satellite.py +1 -3
- tests/numpy_batch/test_sun_position.py +7 -7
- tests/torch_datasets/test_pvnet_uk_regional.py +4 -6
- tests/torch_datasets/test_site.py +85 -0
- ocf_data_sampler-0.0.23.dist-info/RECORD +0 -54
- {ocf_data_sampler-0.0.23.dist-info → ocf_data_sampler-0.0.25.dist-info}/LICENSE +0 -0
ocf_data_sampler/config/model.py
CHANGED
|
@@ -102,6 +102,39 @@ class TimeResolutionMixin(Base):
|
|
|
102
102
|
)
|
|
103
103
|
|
|
104
104
|
|
|
105
|
+
class Site(DataSourceMixin, TimeResolutionMixin, DropoutMixin):
|
|
106
|
+
"""Site configuration model"""
|
|
107
|
+
|
|
108
|
+
file_path: str = Field(
|
|
109
|
+
...,
|
|
110
|
+
description="The NetCDF files holding the power timeseries.",
|
|
111
|
+
)
|
|
112
|
+
metadata_file_path: str = Field(
|
|
113
|
+
...,
|
|
114
|
+
description="The CSV files describing power system",
|
|
115
|
+
)
|
|
116
|
+
|
|
117
|
+
@field_validator("forecast_minutes")
|
|
118
|
+
def forecast_minutes_divide_by_time_resolution(cls, v: int, info: ValidationInfo) -> int:
|
|
119
|
+
"""Check forecast length requested will give stable number of timesteps"""
|
|
120
|
+
if v % info.data["time_resolution_minutes"] != 0:
|
|
121
|
+
message = "Forecast duration must be divisible by time resolution"
|
|
122
|
+
logger.error(message)
|
|
123
|
+
raise Exception(message)
|
|
124
|
+
return v
|
|
125
|
+
|
|
126
|
+
@field_validator("history_minutes")
|
|
127
|
+
def history_minutes_divide_by_time_resolution(cls, v: int, info: ValidationInfo) -> int:
|
|
128
|
+
"""Check history length requested will give stable number of timesteps"""
|
|
129
|
+
if v % info.data["time_resolution_minutes"] != 0:
|
|
130
|
+
message = "History duration must be divisible by time resolution"
|
|
131
|
+
logger.error(message)
|
|
132
|
+
raise Exception(message)
|
|
133
|
+
return v
|
|
134
|
+
|
|
135
|
+
# TODO validate the netcdf for sites
|
|
136
|
+
# TODO validate the csv for metadata
|
|
137
|
+
|
|
105
138
|
class Satellite(DataSourceMixin, TimeResolutionMixin, DropoutMixin):
|
|
106
139
|
"""Satellite configuration model"""
|
|
107
140
|
|
|
@@ -240,6 +273,7 @@ class InputData(Base):
|
|
|
240
273
|
satellite: Optional[Satellite] = None
|
|
241
274
|
nwp: Optional[MultiNWP] = None
|
|
242
275
|
gsp: Optional[GSP] = None
|
|
276
|
+
site: Optional[Site] = None
|
|
243
277
|
|
|
244
278
|
|
|
245
279
|
class Configuration(Base):
|
|
@@ -0,0 +1,55 @@
|
|
|
1
|
+
""" Loads all data sources """
|
|
2
|
+
import xarray as xr
|
|
3
|
+
|
|
4
|
+
from ocf_data_sampler.config import Configuration
|
|
5
|
+
from ocf_data_sampler.load.gsp import open_gsp
|
|
6
|
+
from ocf_data_sampler.load.nwp import open_nwp
|
|
7
|
+
from ocf_data_sampler.load.satellite import open_sat_data
|
|
8
|
+
from ocf_data_sampler.load.site import open_site
|
|
9
|
+
|
|
10
|
+
|
|
11
|
+
def get_dataset_dict(config: Configuration) -> dict[str, dict[xr.DataArray]]:
|
|
12
|
+
"""Construct dictionary of all of the input data sources
|
|
13
|
+
|
|
14
|
+
Args:
|
|
15
|
+
config: Configuration file
|
|
16
|
+
"""
|
|
17
|
+
|
|
18
|
+
in_config = config.input_data
|
|
19
|
+
|
|
20
|
+
datasets_dict = {}
|
|
21
|
+
|
|
22
|
+
# Load GSP data unless the path is None
|
|
23
|
+
if in_config.gsp and in_config.gsp.gsp_zarr_path:
|
|
24
|
+
da_gsp = open_gsp(zarr_path=in_config.gsp.gsp_zarr_path).compute()
|
|
25
|
+
|
|
26
|
+
# Remove national GSP
|
|
27
|
+
datasets_dict["gsp"] = da_gsp.sel(gsp_id=slice(1, None))
|
|
28
|
+
|
|
29
|
+
# Load NWP data if in config
|
|
30
|
+
if in_config.nwp:
|
|
31
|
+
|
|
32
|
+
datasets_dict["nwp"] = {}
|
|
33
|
+
for nwp_source, nwp_config in in_config.nwp.items():
|
|
34
|
+
|
|
35
|
+
da_nwp = open_nwp(nwp_config.nwp_zarr_path, provider=nwp_config.nwp_provider)
|
|
36
|
+
|
|
37
|
+
da_nwp = da_nwp.sel(channel=list(nwp_config.nwp_channels))
|
|
38
|
+
|
|
39
|
+
datasets_dict["nwp"][nwp_source] = da_nwp
|
|
40
|
+
|
|
41
|
+
# Load satellite data if in config
|
|
42
|
+
if in_config.satellite:
|
|
43
|
+
sat_config = config.input_data.satellite
|
|
44
|
+
|
|
45
|
+
da_sat = open_sat_data(sat_config.satellite_zarr_path)
|
|
46
|
+
|
|
47
|
+
da_sat = da_sat.sel(channel=list(sat_config.satellite_channels))
|
|
48
|
+
|
|
49
|
+
datasets_dict["sat"] = da_sat
|
|
50
|
+
|
|
51
|
+
if in_config.site:
|
|
52
|
+
da_sites = open_site(in_config.site)
|
|
53
|
+
datasets_dict["site"] = da_sites
|
|
54
|
+
|
|
55
|
+
return datasets_dict
|
|
@@ -9,7 +9,6 @@ from ocf_data_sampler.load.utils import (
|
|
|
9
9
|
)
|
|
10
10
|
|
|
11
11
|
|
|
12
|
-
|
|
13
12
|
def open_ifs(zarr_path: Path | str | list[Path] | list[str]) -> xr.DataArray:
|
|
14
13
|
"""
|
|
15
14
|
Opens the ECMWF IFS NWP data
|
|
@@ -27,10 +26,14 @@ def open_ifs(zarr_path: Path | str | list[Path] | list[str]) -> xr.DataArray:
|
|
|
27
26
|
ds = ds.rename(
|
|
28
27
|
{
|
|
29
28
|
"init_time": "init_time_utc",
|
|
30
|
-
"variable": "channel",
|
|
31
29
|
}
|
|
32
30
|
)
|
|
33
31
|
|
|
32
|
+
# LEGACY SUPPORT
|
|
33
|
+
# rename variable to channel if it exists
|
|
34
|
+
if "variable" in ds:
|
|
35
|
+
ds = ds.rename({"variable": "channel"})
|
|
36
|
+
|
|
34
37
|
# Check the timestamps are unique and increasing
|
|
35
38
|
check_time_unique_increasing(ds.init_time_utc)
|
|
36
39
|
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
import pandas as pd
|
|
2
|
+
import xarray as xr
|
|
3
|
+
import numpy as np
|
|
4
|
+
|
|
5
|
+
from ocf_data_sampler.config.model import Site
|
|
6
|
+
|
|
7
|
+
|
|
8
|
+
def open_site(sites_config: Site) -> xr.DataArray:
|
|
9
|
+
|
|
10
|
+
# Load site generation xr.Dataset
|
|
11
|
+
data_ds = xr.open_dataset(sites_config.file_path)
|
|
12
|
+
|
|
13
|
+
# Load site generation data
|
|
14
|
+
metadata_df = pd.read_csv(sites_config.metadata_file_path, index_col="site_id")
|
|
15
|
+
|
|
16
|
+
# Add coordinates
|
|
17
|
+
ds = data_ds.assign_coords(
|
|
18
|
+
latitude=(metadata_df.latitude.to_xarray()),
|
|
19
|
+
longitude=(metadata_df.longitude.to_xarray()),
|
|
20
|
+
capacity_kwp=data_ds.capacity_kwp,
|
|
21
|
+
)
|
|
22
|
+
|
|
23
|
+
# Sanity checks
|
|
24
|
+
assert np.isfinite(data_ds.capacity_kwp.values).all()
|
|
25
|
+
assert (data_ds.capacity_kwp.values > 0).all()
|
|
26
|
+
assert metadata_df.index.is_unique
|
|
27
|
+
|
|
28
|
+
return ds.generation_kw
|
|
29
|
+
|
|
30
|
+
|
|
@@ -1,7 +1,8 @@
|
|
|
1
1
|
"""Conversion from Xarray to NumpyBatch"""
|
|
2
2
|
|
|
3
|
-
from .gsp import convert_gsp_to_numpy_batch
|
|
4
|
-
from .nwp import convert_nwp_to_numpy_batch
|
|
5
|
-
from .satellite import convert_satellite_to_numpy_batch
|
|
3
|
+
from .gsp import convert_gsp_to_numpy_batch, GSPBatchKey
|
|
4
|
+
from .nwp import convert_nwp_to_numpy_batch, NWPBatchKey
|
|
5
|
+
from .satellite import convert_satellite_to_numpy_batch, SatelliteBatchKey
|
|
6
6
|
from .sun_position import make_sun_position_numpy_batch
|
|
7
|
+
from .site import convert_site_to_numpy_batch
|
|
7
8
|
|
|
@@ -6,15 +6,15 @@ import xarray as xr
|
|
|
6
6
|
class GSPBatchKey:
|
|
7
7
|
|
|
8
8
|
gsp = 'gsp'
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
9
|
+
nominal_capacity_mwp = 'gsp_nominal_capacity_mwp'
|
|
10
|
+
effective_capacity_mwp = 'gsp_effective_capacity_mwp'
|
|
11
|
+
time_utc = 'gsp_time_utc'
|
|
12
|
+
t0_idx = 'gsp_t0_idx'
|
|
13
|
+
solar_azimuth = 'gsp_solar_azimuth'
|
|
14
|
+
solar_elevation = 'gsp_solar_elevation'
|
|
15
15
|
gsp_id = 'gsp_id'
|
|
16
|
-
|
|
17
|
-
|
|
16
|
+
x_osgb = 'gsp_x_osgb'
|
|
17
|
+
y_osgb = 'gsp_y_osgb'
|
|
18
18
|
|
|
19
19
|
|
|
20
20
|
def convert_gsp_to_numpy_batch(da: xr.DataArray, t0_idx: int | None = None) -> dict:
|
|
@@ -22,12 +22,12 @@ def convert_gsp_to_numpy_batch(da: xr.DataArray, t0_idx: int | None = None) -> d
|
|
|
22
22
|
|
|
23
23
|
example = {
|
|
24
24
|
GSPBatchKey.gsp: da.values,
|
|
25
|
-
GSPBatchKey.
|
|
26
|
-
GSPBatchKey.
|
|
27
|
-
GSPBatchKey.
|
|
25
|
+
GSPBatchKey.nominal_capacity_mwp: da.isel(time_utc=0)["nominal_capacity_mwp"].values,
|
|
26
|
+
GSPBatchKey.effective_capacity_mwp: da.isel(time_utc=0)["effective_capacity_mwp"].values,
|
|
27
|
+
GSPBatchKey.time_utc: da["time_utc"].values.astype(float),
|
|
28
28
|
}
|
|
29
29
|
|
|
30
30
|
if t0_idx is not None:
|
|
31
|
-
example[GSPBatchKey.
|
|
31
|
+
example[GSPBatchKey.t0_idx] = t0_idx
|
|
32
32
|
|
|
33
33
|
return example
|
|
@@ -7,13 +7,13 @@ import xarray as xr
|
|
|
7
7
|
class NWPBatchKey:
|
|
8
8
|
|
|
9
9
|
nwp = 'nwp'
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
10
|
+
channel_names = 'nwp_channel_names'
|
|
11
|
+
init_time_utc = 'nwp_init_time_utc'
|
|
12
|
+
step = 'nwp_step'
|
|
13
|
+
target_time_utc = 'nwp_target_time_utc'
|
|
14
|
+
t0_idx = 'nwp_t0_idx'
|
|
15
|
+
y_osgb = 'nwp_y_osgb'
|
|
16
|
+
x_osgb = 'nwp_x_osgb'
|
|
17
17
|
|
|
18
18
|
|
|
19
19
|
def convert_nwp_to_numpy_batch(da: xr.DataArray, t0_idx: int | None = None) -> dict:
|
|
@@ -21,23 +21,23 @@ def convert_nwp_to_numpy_batch(da: xr.DataArray, t0_idx: int | None = None) -> d
|
|
|
21
21
|
|
|
22
22
|
example = {
|
|
23
23
|
NWPBatchKey.nwp: da.values,
|
|
24
|
-
NWPBatchKey.
|
|
25
|
-
NWPBatchKey.
|
|
26
|
-
NWPBatchKey.
|
|
24
|
+
NWPBatchKey.channel_names: da.channel.values,
|
|
25
|
+
NWPBatchKey.init_time_utc: da.init_time_utc.values.astype(float),
|
|
26
|
+
NWPBatchKey.step: (da.step.values / pd.Timedelta("1h")).astype(int),
|
|
27
27
|
}
|
|
28
28
|
|
|
29
29
|
if "target_time_utc" in da.coords:
|
|
30
|
-
example[NWPBatchKey.
|
|
30
|
+
example[NWPBatchKey.target_time_utc] = da.target_time_utc.values.astype(float)
|
|
31
31
|
|
|
32
32
|
# TODO: Do we need this at all? Especially since it is only present in UKV data
|
|
33
33
|
for batch_key, dataset_key in (
|
|
34
|
-
(NWPBatchKey.
|
|
35
|
-
(NWPBatchKey.
|
|
34
|
+
(NWPBatchKey.y_osgb, "y_osgb"),
|
|
35
|
+
(NWPBatchKey.x_osgb, "x_osgb"),
|
|
36
36
|
):
|
|
37
37
|
if dataset_key in da.coords:
|
|
38
38
|
example[batch_key] = da[dataset_key].values
|
|
39
39
|
|
|
40
40
|
if t0_idx is not None:
|
|
41
|
-
example[NWPBatchKey.
|
|
41
|
+
example[NWPBatchKey.t0_idx] = t0_idx
|
|
42
42
|
|
|
43
43
|
return example
|
|
@@ -5,26 +5,26 @@ import xarray as xr
|
|
|
5
5
|
class SatelliteBatchKey:
|
|
6
6
|
|
|
7
7
|
satellite_actual = 'satellite_actual'
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
|
|
8
|
+
time_utc = 'satellite_time_utc'
|
|
9
|
+
x_geostationary = 'satellite_x_geostationary'
|
|
10
|
+
y_geostationary = 'satellite_y_geostationary'
|
|
11
|
+
t0_idx = 'satellite_t0_idx'
|
|
12
12
|
|
|
13
13
|
|
|
14
14
|
def convert_satellite_to_numpy_batch(da: xr.DataArray, t0_idx: int | None = None) -> dict:
|
|
15
15
|
"""Convert from Xarray to NumpyBatch"""
|
|
16
16
|
example = {
|
|
17
17
|
SatelliteBatchKey.satellite_actual: da.values,
|
|
18
|
-
SatelliteBatchKey.
|
|
18
|
+
SatelliteBatchKey.time_utc: da.time_utc.values.astype(float),
|
|
19
19
|
}
|
|
20
20
|
|
|
21
21
|
for batch_key, dataset_key in (
|
|
22
|
-
(SatelliteBatchKey.
|
|
23
|
-
(SatelliteBatchKey.
|
|
22
|
+
(SatelliteBatchKey.x_geostationary, "x_geostationary"),
|
|
23
|
+
(SatelliteBatchKey.y_geostationary, "y_geostationary"),
|
|
24
24
|
):
|
|
25
25
|
example[batch_key] = da[dataset_key].values
|
|
26
26
|
|
|
27
27
|
if t0_idx is not None:
|
|
28
|
-
example[SatelliteBatchKey.
|
|
28
|
+
example[SatelliteBatchKey.t0_idx] = t0_idx
|
|
29
29
|
|
|
30
30
|
return example
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
"""Convert site to Numpy Batch"""
|
|
2
|
+
|
|
3
|
+
import xarray as xr
|
|
4
|
+
|
|
5
|
+
|
|
6
|
+
class SiteBatchKey:
|
|
7
|
+
|
|
8
|
+
generation = "site"
|
|
9
|
+
site_capacity_kwp = "site_capacity_kwp"
|
|
10
|
+
site_time_utc = "site_time_utc"
|
|
11
|
+
site_t0_idx = "site_t0_idx"
|
|
12
|
+
site_solar_azimuth = "site_solar_azimuth"
|
|
13
|
+
site_solar_elevation = "site_solar_elevation"
|
|
14
|
+
site_id = "site_id"
|
|
15
|
+
|
|
16
|
+
|
|
17
|
+
def convert_site_to_numpy_batch(da: xr.DataArray, t0_idx: int | None = None) -> dict:
|
|
18
|
+
"""Convert from Xarray to NumpyBatch"""
|
|
19
|
+
|
|
20
|
+
example = {
|
|
21
|
+
SiteBatchKey.generation: da.values,
|
|
22
|
+
SiteBatchKey.site_capacity_kwp: da.isel(time_utc=0)["capacity_kwp"].values,
|
|
23
|
+
SiteBatchKey.site_time_utc: da["time_utc"].values.astype(float),
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
if t0_idx is not None:
|
|
27
|
+
example[SiteBatchKey.site_t0_idx] = t0_idx
|
|
28
|
+
|
|
29
|
+
return example
|
|
@@ -1 +1,8 @@
|
|
|
1
|
-
|
|
1
|
+
from .fill_time_periods import fill_time_periods
|
|
2
|
+
from .find_contiguous_time_periods import (
|
|
3
|
+
find_contiguous_t0_periods,
|
|
4
|
+
intersection_of_multiple_dataframes_of_periods,
|
|
5
|
+
)
|
|
6
|
+
from .location import Location
|
|
7
|
+
from .spatial_slice_for_dataset import slice_datasets_by_space
|
|
8
|
+
from .time_slice_for_dataset import slice_datasets_by_time
|
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
""" Functions for simulating dropout in time series data """
|
|
1
2
|
import numpy as np
|
|
2
3
|
import pandas as pd
|
|
3
4
|
import xarray as xr
|
|
@@ -5,7 +6,7 @@ import xarray as xr
|
|
|
5
6
|
|
|
6
7
|
def draw_dropout_time(
|
|
7
8
|
t0: pd.Timestamp,
|
|
8
|
-
dropout_timedeltas: list[pd.Timedelta] | None,
|
|
9
|
+
dropout_timedeltas: list[pd.Timedelta] | pd.Timedelta | None,
|
|
9
10
|
dropout_frac: float = 0,
|
|
10
11
|
):
|
|
11
12
|
|
|
@@ -55,6 +55,23 @@ def lon_lat_to_osgb(
|
|
|
55
55
|
return _lon_lat_to_osgb(xx=x, yy=y)
|
|
56
56
|
|
|
57
57
|
|
|
58
|
+
def lon_lat_to_geostationary_area_coords(
|
|
59
|
+
longitude: Union[Number, np.ndarray],
|
|
60
|
+
latitude: Union[Number, np.ndarray],
|
|
61
|
+
xr_data: xr.DataArray,
|
|
62
|
+
) -> tuple[Union[Number, np.ndarray], Union[Number, np.ndarray]]:
|
|
63
|
+
"""Loads geostationary area and transformation from lat-lon to geostationary coords
|
|
64
|
+
|
|
65
|
+
Args:
|
|
66
|
+
longitude: longitude
|
|
67
|
+
latitude: latitude
|
|
68
|
+
xr_data: xarray object with geostationary area
|
|
69
|
+
|
|
70
|
+
Returns:
|
|
71
|
+
Geostationary coords: x, y
|
|
72
|
+
"""
|
|
73
|
+
return coordinates_to_geostationary_area_coords(longitude, latitude, xr_data, WGS84)
|
|
74
|
+
|
|
58
75
|
def osgb_to_geostationary_area_coords(
|
|
59
76
|
x: Union[Number, np.ndarray],
|
|
60
77
|
y: Union[Number, np.ndarray],
|
|
@@ -70,6 +87,31 @@ def osgb_to_geostationary_area_coords(
|
|
|
70
87
|
Returns:
|
|
71
88
|
Geostationary coords: x, y
|
|
72
89
|
"""
|
|
90
|
+
|
|
91
|
+
return coordinates_to_geostationary_area_coords(x, y, xr_data, OSGB36)
|
|
92
|
+
|
|
93
|
+
|
|
94
|
+
|
|
95
|
+
def coordinates_to_geostationary_area_coords(
|
|
96
|
+
x: Union[Number, np.ndarray],
|
|
97
|
+
y: Union[Number, np.ndarray],
|
|
98
|
+
xr_data: xr.DataArray,
|
|
99
|
+
crs_from: int
|
|
100
|
+
) -> tuple[Union[Number, np.ndarray], Union[Number, np.ndarray]]:
|
|
101
|
+
"""Loads geostationary area and transformation from respective coordiates to geostationary coords
|
|
102
|
+
|
|
103
|
+
Args:
|
|
104
|
+
x: osgb east-west, or latitude
|
|
105
|
+
y: osgb north-south, or longitude
|
|
106
|
+
xr_data: xarray object with geostationary area
|
|
107
|
+
crs_from: the cordiates system of x,y
|
|
108
|
+
|
|
109
|
+
Returns:
|
|
110
|
+
Geostationary coords: x, y
|
|
111
|
+
"""
|
|
112
|
+
|
|
113
|
+
assert crs_from in [OSGB36, WGS84], f"Unrecognized coordinate system: {crs_from}"
|
|
114
|
+
|
|
73
115
|
# Only load these if using geostationary projection
|
|
74
116
|
import pyresample
|
|
75
117
|
|
|
@@ -80,7 +122,7 @@ def osgb_to_geostationary_area_coords(
|
|
|
80
122
|
)
|
|
81
123
|
geostationary_crs = geostationary_area_definition.crs
|
|
82
124
|
osgb_to_geostationary = pyproj.Transformer.from_crs(
|
|
83
|
-
crs_from=
|
|
125
|
+
crs_from=crs_from, crs_to=geostationary_crs, always_xy=True
|
|
84
126
|
).transform
|
|
85
127
|
return osgb_to_geostationary(xx=x, yy=y)
|
|
86
128
|
|
|
@@ -8,6 +8,7 @@ import xarray as xr
|
|
|
8
8
|
from ocf_data_sampler.select.location import Location
|
|
9
9
|
from ocf_data_sampler.select.geospatial import (
|
|
10
10
|
lon_lat_to_osgb,
|
|
11
|
+
lon_lat_to_geostationary_area_coords,
|
|
11
12
|
osgb_to_geostationary_area_coords,
|
|
12
13
|
osgb_to_lon_lat,
|
|
13
14
|
spatial_coord_type,
|
|
@@ -101,7 +102,7 @@ def _get_idx_of_pixel_closest_to_poi(
|
|
|
101
102
|
|
|
102
103
|
def _get_idx_of_pixel_closest_to_poi_geostationary(
|
|
103
104
|
da: xr.DataArray,
|
|
104
|
-
|
|
105
|
+
center: Location,
|
|
105
106
|
) -> Location:
|
|
106
107
|
"""
|
|
107
108
|
Return x and y index location of pixel at center of region of interest.
|
|
@@ -116,7 +117,12 @@ def _get_idx_of_pixel_closest_to_poi_geostationary(
|
|
|
116
117
|
|
|
117
118
|
_, x_dim, y_dim = spatial_coord_type(da)
|
|
118
119
|
|
|
119
|
-
|
|
120
|
+
if center.coordinate_system == 'osgb':
|
|
121
|
+
x, y = osgb_to_geostationary_area_coords(x=center.x, y=center.y, xr_data=da)
|
|
122
|
+
elif center.coordinate_system == 'lon_lat':
|
|
123
|
+
x, y = lon_lat_to_geostationary_area_coords(longitude=center.x, latitude=center.y, xr_data=da)
|
|
124
|
+
else:
|
|
125
|
+
x,y = center.x, center.y
|
|
120
126
|
center_geostationary = Location(x=x, y=y, coordinate_system="geostationary")
|
|
121
127
|
|
|
122
128
|
# Check that the requested point lies within the data
|
|
@@ -0,0 +1,53 @@
|
|
|
1
|
+
""" Functions for selecting data around a given location """
|
|
2
|
+
from ocf_data_sampler.config import Configuration
|
|
3
|
+
from ocf_data_sampler.select.location import Location
|
|
4
|
+
from ocf_data_sampler.select.select_spatial_slice import select_spatial_slice_pixels
|
|
5
|
+
|
|
6
|
+
|
|
7
|
+
def slice_datasets_by_space(
|
|
8
|
+
datasets_dict: dict,
|
|
9
|
+
location: Location,
|
|
10
|
+
config: Configuration,
|
|
11
|
+
) -> dict:
|
|
12
|
+
"""Slice the dictionary of input data sources around a given location
|
|
13
|
+
|
|
14
|
+
Args:
|
|
15
|
+
datasets_dict: Dictionary of the input data sources
|
|
16
|
+
location: The location to sample around
|
|
17
|
+
config: Configuration object.
|
|
18
|
+
"""
|
|
19
|
+
|
|
20
|
+
assert set(datasets_dict.keys()).issubset({"nwp", "sat", "gsp", "site"})
|
|
21
|
+
|
|
22
|
+
sliced_datasets_dict = {}
|
|
23
|
+
|
|
24
|
+
if "nwp" in datasets_dict:
|
|
25
|
+
|
|
26
|
+
sliced_datasets_dict["nwp"] = {}
|
|
27
|
+
|
|
28
|
+
for nwp_key, nwp_config in config.input_data.nwp.items():
|
|
29
|
+
|
|
30
|
+
sliced_datasets_dict["nwp"][nwp_key] = select_spatial_slice_pixels(
|
|
31
|
+
datasets_dict["nwp"][nwp_key],
|
|
32
|
+
location,
|
|
33
|
+
height_pixels=nwp_config.nwp_image_size_pixels_height,
|
|
34
|
+
width_pixels=nwp_config.nwp_image_size_pixels_width,
|
|
35
|
+
)
|
|
36
|
+
|
|
37
|
+
if "sat" in datasets_dict:
|
|
38
|
+
sat_config = config.input_data.satellite
|
|
39
|
+
|
|
40
|
+
sliced_datasets_dict["sat"] = select_spatial_slice_pixels(
|
|
41
|
+
datasets_dict["sat"],
|
|
42
|
+
location,
|
|
43
|
+
height_pixels=sat_config.satellite_image_size_pixels_height,
|
|
44
|
+
width_pixels=sat_config.satellite_image_size_pixels_width,
|
|
45
|
+
)
|
|
46
|
+
|
|
47
|
+
if "gsp" in datasets_dict:
|
|
48
|
+
sliced_datasets_dict["gsp"] = datasets_dict["gsp"].sel(gsp_id=location.id)
|
|
49
|
+
|
|
50
|
+
if "site" in datasets_dict:
|
|
51
|
+
sliced_datasets_dict["site"] = datasets_dict["site"].sel(site_id=location.id)
|
|
52
|
+
|
|
53
|
+
return sliced_datasets_dict
|
|
@@ -0,0 +1,124 @@
|
|
|
1
|
+
""" Slice datasets by time"""
|
|
2
|
+
import pandas as pd
|
|
3
|
+
|
|
4
|
+
from ocf_data_sampler.config import Configuration
|
|
5
|
+
from ocf_data_sampler.select.dropout import draw_dropout_time, apply_dropout_time
|
|
6
|
+
from ocf_data_sampler.select.select_time_slice import select_time_slice_nwp, select_time_slice
|
|
7
|
+
from ocf_data_sampler.time_functions import minutes
|
|
8
|
+
|
|
9
|
+
|
|
10
|
+
def slice_datasets_by_time(
|
|
11
|
+
datasets_dict: dict,
|
|
12
|
+
t0: pd.Timestamp,
|
|
13
|
+
config: Configuration,
|
|
14
|
+
) -> dict:
|
|
15
|
+
"""Slice the dictionary of input data sources around a given t0 time
|
|
16
|
+
|
|
17
|
+
Args:
|
|
18
|
+
datasets_dict: Dictionary of the input data sources
|
|
19
|
+
t0: The init-time
|
|
20
|
+
config: Configuration object.
|
|
21
|
+
"""
|
|
22
|
+
|
|
23
|
+
sliced_datasets_dict = {}
|
|
24
|
+
|
|
25
|
+
if "nwp" in datasets_dict:
|
|
26
|
+
|
|
27
|
+
sliced_datasets_dict["nwp"] = {}
|
|
28
|
+
|
|
29
|
+
for nwp_key, da_nwp in datasets_dict["nwp"].items():
|
|
30
|
+
|
|
31
|
+
nwp_config = config.input_data.nwp[nwp_key]
|
|
32
|
+
|
|
33
|
+
sliced_datasets_dict["nwp"][nwp_key] = select_time_slice_nwp(
|
|
34
|
+
da_nwp,
|
|
35
|
+
t0,
|
|
36
|
+
sample_period_duration=minutes(nwp_config.time_resolution_minutes),
|
|
37
|
+
history_duration=minutes(nwp_config.history_minutes),
|
|
38
|
+
forecast_duration=minutes(nwp_config.forecast_minutes),
|
|
39
|
+
dropout_timedeltas=minutes(nwp_config.dropout_timedeltas_minutes),
|
|
40
|
+
dropout_frac=nwp_config.dropout_fraction,
|
|
41
|
+
accum_channels=nwp_config.nwp_accum_channels,
|
|
42
|
+
)
|
|
43
|
+
|
|
44
|
+
if "sat" in datasets_dict:
|
|
45
|
+
|
|
46
|
+
sat_config = config.input_data.satellite
|
|
47
|
+
|
|
48
|
+
sliced_datasets_dict["sat"] = select_time_slice(
|
|
49
|
+
datasets_dict["sat"],
|
|
50
|
+
t0,
|
|
51
|
+
sample_period_duration=minutes(sat_config.time_resolution_minutes),
|
|
52
|
+
interval_start=minutes(-sat_config.history_minutes),
|
|
53
|
+
interval_end=minutes(-sat_config.live_delay_minutes),
|
|
54
|
+
max_steps_gap=2,
|
|
55
|
+
)
|
|
56
|
+
|
|
57
|
+
# Randomly sample dropout
|
|
58
|
+
sat_dropout_time = draw_dropout_time(
|
|
59
|
+
t0,
|
|
60
|
+
dropout_timedeltas=minutes(sat_config.dropout_timedeltas_minutes),
|
|
61
|
+
dropout_frac=sat_config.dropout_fraction,
|
|
62
|
+
)
|
|
63
|
+
|
|
64
|
+
# Apply the dropout
|
|
65
|
+
sliced_datasets_dict["sat"] = apply_dropout_time(
|
|
66
|
+
sliced_datasets_dict["sat"],
|
|
67
|
+
sat_dropout_time,
|
|
68
|
+
)
|
|
69
|
+
|
|
70
|
+
if "gsp" in datasets_dict:
|
|
71
|
+
gsp_config = config.input_data.gsp
|
|
72
|
+
|
|
73
|
+
sliced_datasets_dict["gsp_future"] = select_time_slice(
|
|
74
|
+
datasets_dict["gsp"],
|
|
75
|
+
t0,
|
|
76
|
+
sample_period_duration=minutes(gsp_config.time_resolution_minutes),
|
|
77
|
+
interval_start=minutes(30),
|
|
78
|
+
interval_end=minutes(gsp_config.forecast_minutes),
|
|
79
|
+
)
|
|
80
|
+
|
|
81
|
+
sliced_datasets_dict["gsp"] = select_time_slice(
|
|
82
|
+
datasets_dict["gsp"],
|
|
83
|
+
t0,
|
|
84
|
+
sample_period_duration=minutes(gsp_config.time_resolution_minutes),
|
|
85
|
+
interval_start=-minutes(gsp_config.history_minutes),
|
|
86
|
+
interval_end=minutes(0),
|
|
87
|
+
)
|
|
88
|
+
|
|
89
|
+
# Dropout on the GSP, but not the future GSP
|
|
90
|
+
gsp_dropout_time = draw_dropout_time(
|
|
91
|
+
t0,
|
|
92
|
+
dropout_timedeltas=minutes(gsp_config.dropout_timedeltas_minutes),
|
|
93
|
+
dropout_frac=gsp_config.dropout_fraction,
|
|
94
|
+
)
|
|
95
|
+
|
|
96
|
+
sliced_datasets_dict["gsp"] = apply_dropout_time(
|
|
97
|
+
sliced_datasets_dict["gsp"], gsp_dropout_time
|
|
98
|
+
)
|
|
99
|
+
|
|
100
|
+
if "site" in datasets_dict:
|
|
101
|
+
site_config = config.input_data.site
|
|
102
|
+
|
|
103
|
+
sliced_datasets_dict["site"] = select_time_slice(
|
|
104
|
+
datasets_dict["site"],
|
|
105
|
+
t0,
|
|
106
|
+
sample_period_duration=minutes(site_config.time_resolution_minutes),
|
|
107
|
+
interval_start=-minutes(site_config.history_minutes),
|
|
108
|
+
interval_end=minutes(site_config.forecast_minutes),
|
|
109
|
+
)
|
|
110
|
+
|
|
111
|
+
# Randomly sample dropout
|
|
112
|
+
site_dropout_time = draw_dropout_time(
|
|
113
|
+
t0,
|
|
114
|
+
dropout_timedeltas=minutes(site_config.dropout_timedeltas_minutes),
|
|
115
|
+
dropout_frac=site_config.dropout_fraction,
|
|
116
|
+
)
|
|
117
|
+
|
|
118
|
+
# Apply the dropout
|
|
119
|
+
sliced_datasets_dict["site"] = apply_dropout_time(
|
|
120
|
+
sliced_datasets_dict["site"],
|
|
121
|
+
site_dropout_time,
|
|
122
|
+
)
|
|
123
|
+
|
|
124
|
+
return sliced_datasets_dict
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
import pandas as pd
|
|
2
|
+
|
|
3
|
+
|
|
4
|
+
def minutes(minutes: int | list[float]) -> pd.Timedelta | pd.TimedeltaIndex:
|
|
5
|
+
"""Timedelta minutes
|
|
6
|
+
|
|
7
|
+
Args:
|
|
8
|
+
minutes: the number of minutes, single value or list
|
|
9
|
+
"""
|
|
10
|
+
minutes_delta = pd.to_timedelta(minutes, unit="m")
|
|
11
|
+
return minutes_delta
|