diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index f4d042f003042319b3867857b756665a2aa3ddfc..eacbe3e26323e0a0bf1579cba53e2e12ecfd27c0 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -42,7 +42,7 @@ tests (from scratch):
- ./CI/update_badge.sh > /dev/null
script:
- pip install --upgrade pip
- - pip install numpy wheel six
+ - pip install numpy wheel six==1.15.0
- zypper --non-interactive install binutils libproj-devel gdal-devel
- zypper --non-interactive install proj geos-devel
# - cat requirements.txt | cut -f1 -d"#" | sed '/^\s*$/d' | xargs -L 1 pip install
diff --git a/HPC_setup/requirements_HDFML_additionals.txt b/HPC_setup/requirements_HDFML_additionals.txt
index 12e09ccdd620c0c81c78ae6d4781d4feb5b94baf..b2a29fbfb353f24d8c99d8429693022ea1fd406f 100644
--- a/HPC_setup/requirements_HDFML_additionals.txt
+++ b/HPC_setup/requirements_HDFML_additionals.txt
@@ -2,6 +2,7 @@ absl-py==0.11.0
appdirs==1.4.4
astor==0.8.1
attrs==20.3.0
+bottleneck==1.3.2
cached-property==1.5.2
certifi==2020.12.5
cftime==1.4.1
@@ -9,6 +10,7 @@ chardet==4.0.0
coverage==5.4
cycler==0.10.0
dask==2021.2.0
+dill==0.3.3
fsspec==0.8.5
gast==0.4.0
grpcio==1.35.0
diff --git a/HPC_setup/requirements_JUWELS_additionals.txt b/HPC_setup/requirements_JUWELS_additionals.txt
index 12e09ccdd620c0c81c78ae6d4781d4feb5b94baf..b2a29fbfb353f24d8c99d8429693022ea1fd406f 100644
--- a/HPC_setup/requirements_JUWELS_additionals.txt
+++ b/HPC_setup/requirements_JUWELS_additionals.txt
@@ -2,6 +2,7 @@ absl-py==0.11.0
appdirs==1.4.4
astor==0.8.1
attrs==20.3.0
+bottleneck==1.3.2
cached-property==1.5.2
certifi==2020.12.5
cftime==1.4.1
@@ -9,6 +10,7 @@ chardet==4.0.0
coverage==5.4
cycler==0.10.0
dask==2021.2.0
+dill==0.3.3
fsspec==0.8.5
gast==0.4.0
grpcio==1.35.0
diff --git a/mlair/data_handler/abstract_data_handler.py b/mlair/data_handler/abstract_data_handler.py
index f085d18bb8d33839a0e3b5f6f3d5ada92134e7f6..419db059a58beeb4ed7e3e198e41b565f8dc7d25 100644
--- a/mlair/data_handler/abstract_data_handler.py
+++ b/mlair/data_handler/abstract_data_handler.py
@@ -55,3 +55,6 @@ class AbstractDataHandler:
def get_coordinates(self) -> Union[None, Dict]:
"""Return coordinates as dictionary with keys `lon` and `lat`."""
return None
+
+ def _hash_list(self):
+ return []
diff --git a/mlair/data_handler/data_handler_kz_filter.py b/mlair/data_handler/data_handler_kz_filter.py
index 78638a13b4ea50cd073ca4599a291342fad849d4..539712b39e51c32203e1c55e28ce2eff24069479 100644
--- a/mlair/data_handler/data_handler_kz_filter.py
+++ b/mlair/data_handler/data_handler_kz_filter.py
@@ -7,7 +7,8 @@ import inspect
import numpy as np
import pandas as pd
import xarray as xr
-from typing import List, Union
+from typing import List, Union, Tuple, Optional
+from functools import partial
from mlair.data_handler.data_handler_single_station import DataHandlerSingleStation
from mlair.data_handler import DefaultDataHandler
@@ -22,6 +23,7 @@ class DataHandlerKzFilterSingleStation(DataHandlerSingleStation):
"""Data handler for a single station to be used by a superior data handler. Inputs are kz filtered."""
_requirements = remove_items(inspect.getfullargspec(DataHandlerSingleStation).args, ["self", "station"])
+ _hash = DataHandlerSingleStation._hash + ["kz_filter_length", "kz_filter_iter", "filter_dim"]
DEFAULT_FILTER_DIM = "filter"
@@ -35,13 +37,23 @@ class DataHandlerKzFilterSingleStation(DataHandlerSingleStation):
self.cutoff_period_days = None
super().__init__(*args, **kwargs)
+ def setup_transformation(self, transformation: Union[None, dict, Tuple]) -> Tuple[Optional[dict], Optional[dict]]:
+ """
+ Adjust setup of transformation because kfz filtered data will have negative values which is not compatible with
+ the log transformation. Therefore, replace all log transformation methods by a default standardization. This is
+ only applied on input side.
+ """
+ transformation = super(__class__, self).setup_transformation(transformation)
+ if transformation[0] is not None:
+ for k, v in transformation[0].items():
+ if v["method"] == "log":
+ transformation[0][k]["method"] = "standardise"
+ return transformation
+
def _check_sampling(self, **kwargs):
assert kwargs.get("sampling") == "hourly" # This data handler requires hourly data resolution
- def setup_samples(self):
- """
- Setup samples. This method prepares and creates samples X, and labels Y.
- """
+ def make_input_target(self):
data, self.meta = self.load_data(self.path, self.station, self.statistics_per_var, self.sampling,
self.station_type, self.network, self.store_data_locally, self.data_origin)
self._data = self.interpolate(data, dim=self.time_dim, method=self.interpolation_method,
@@ -54,9 +66,6 @@ class DataHandlerKzFilterSingleStation(DataHandlerSingleStation):
# import matplotlib.pyplot as plt
# self.input_data.sel(filter="74d", variables="temp", Stations="DEBW107").plot()
# self.input_data.sel(variables="temp", Stations="DEBW107").plot.line(hue="filter")
- if self.do_transformation is True:
- self.call_transform()
- self.make_samples()
@TimeTrackingWrapper
def apply_kz_filter(self):
@@ -88,6 +97,15 @@ class DataHandlerKzFilterSingleStation(DataHandlerSingleStation):
return self.history.transpose(self.time_dim, self.window_dim, self.iter_dim, self.target_dim,
self.filter_dim).copy()
+ def _create_lazy_data(self):
+ return [self._data, self.meta, self.input_data, self.target_data, self.cutoff_period, self.cutoff_period_days]
+
+ def _extract_lazy(self, lazy_data):
+ _data, self.meta, _input_data, _target_data, self.cutoff_period, self.cutoff_period_days = lazy_data
+ f_prep = partial(self._slice_prep, start=self.start, end=self.end)
+ self._data, self.input_data, self.target_data = list(map(f_prep, [_data, _input_data, _target_data]))
+
+
class DataHandlerKzFilter(DefaultDataHandler):
"""Data handler using kz filtered data."""
diff --git a/mlair/data_handler/data_handler_mixed_sampling.py b/mlair/data_handler/data_handler_mixed_sampling.py
index caaa7a62d1b772808dcaf58abdfa5483e80861e7..75e9e64506231f32406934b67e65454d87a43f61 100644
--- a/mlair/data_handler/data_handler_mixed_sampling.py
+++ b/mlair/data_handler/data_handler_mixed_sampling.py
@@ -12,6 +12,7 @@ import inspect
from typing import Callable
import datetime as dt
from typing import Any
+from functools import partial
import numpy as np
import pandas as pd
@@ -54,15 +55,9 @@ class DataHandlerMixedSamplingSingleStation(DataHandlerSingleStation):
assert len(parameter) == 2 # (inputs, targets)
kwargs.update({parameter_name: parameter})
- def setup_samples(self):
- """
- Setup samples. This method prepares and creates samples X, and labels Y.
- """
+ def make_input_target(self):
self._data = list(map(self.load_and_interpolate, [0, 1])) # load input (0) and target (1) data
self.set_inputs_and_targets()
- if self.do_transformation is True:
- self.call_transform()
- self.make_samples()
def load_and_interpolate(self, ind) -> [xr.DataArray, pd.DataFrame]:
vars = [self.variables, self.target_var]
@@ -83,6 +78,12 @@ class DataHandlerMixedSamplingSingleStation(DataHandlerSingleStation):
assert len(sampling) == 2
return list(map(lambda x: super(__class__, self).setup_data_path(data_path, x), sampling))
+ def _extract_lazy(self, lazy_data):
+ _data, self.meta, _input_data, _target_data = lazy_data
+ f_prep = partial(self._slice_prep, start=self.start, end=self.end)
+ self._data = f_prep(_data[0]), f_prep(_data[1])
+ self.input_data, self.target_data = list(map(f_prep, [_input_data, _target_data]))
+
class DataHandlerMixedSampling(DefaultDataHandler):
"""Data handler using mixed sampling for input and target."""
@@ -104,19 +105,14 @@ class DataHandlerMixedSamplingWithFilterSingleStation(DataHandlerMixedSamplingSi
def _check_sampling(self, **kwargs):
assert kwargs.get("sampling") == ("hourly", "daily")
- def setup_samples(self):
+ def make_input_target(self):
"""
- Setup samples. This method prepares and creates samples X, and labels Y.
-
A KZ filter is applied on the input data that has hourly resolution. Lables Y are provided as aggregated values
with daily resolution.
"""
self._data = list(map(self.load_and_interpolate, [0, 1])) # load input (0) and target (1) data
self.set_inputs_and_targets()
self.apply_kz_filter()
- if self.do_transformation is True:
- self.call_transform()
- self.make_samples()
def estimate_filter_width(self):
"""
@@ -130,14 +126,24 @@ class DataHandlerMixedSamplingWithFilterSingleStation(DataHandlerMixedSamplingSi
new_date = dt.datetime.strptime(date, "%Y-%m-%d") + dt.timedelta(hours=delta)
return new_date.strftime("%Y-%m-%d")
- def load_and_interpolate(self, ind) -> [xr.DataArray, pd.DataFrame]:
-
+ def update_start_end(self, ind):
if ind == 0: # for inputs
estimated_filter_width = self.estimate_filter_width()
start = self._add_time_delta(self.start, -estimated_filter_width)
end = self._add_time_delta(self.end, estimated_filter_width)
else: # target
start, end = self.start, self.end
+ return start, end
+
+ def load_and_interpolate(self, ind) -> [xr.DataArray, pd.DataFrame]:
+
+ start, end = self.update_start_end(ind)
+ # if ind == 0: # for inputs
+ # estimated_filter_width = self.estimate_filter_width()
+ # start = self._add_time_delta(self.start, -estimated_filter_width)
+ # end = self._add_time_delta(self.end, estimated_filter_width)
+ # else: # target
+ # start, end = self.start, self.end
vars = [self.variables, self.target_var]
stats_per_var = helpers.select_from_dict(self.statistics_per_var, vars[ind])
@@ -149,6 +155,13 @@ class DataHandlerMixedSamplingWithFilterSingleStation(DataHandlerMixedSamplingSi
limit=self.interpolation_limit[ind])
return data
+ def _extract_lazy(self, lazy_data):
+ _data, self.meta, _input_data, _target_data, self.cutoff_period, self.cutoff_period_days = lazy_data
+ start_inp, end_inp = self.update_start_end(0)
+ self._data = list(map(lambda x: self._slice_prep(_data[x], *self.update_start_end(x)), [0, 1]))
+ self.input_data = self._slice_prep(_input_data, start_inp, end_inp)
+ self.target_data = self._slice_prep(_target_data, self.start, self.end)
+
class DataHandlerMixedSamplingWithFilter(DefaultDataHandler):
"""Data handler using mixed sampling for input and target. Inputs are temporal filtered."""
@@ -169,6 +182,7 @@ class DataHandlerSeparationOfScalesSingleStation(DataHandlerMixedSamplingWithFil
"""
_requirements = DataHandlerMixedSamplingWithFilterSingleStation.requirements()
+ _hash = DataHandlerMixedSamplingWithFilterSingleStation._hash + ["time_delta"]
def __init__(self, *args, time_delta=np.sqrt, **kwargs):
assert isinstance(time_delta, Callable)
@@ -204,7 +218,7 @@ class DataHandlerSeparationOfScalesSingleStation(DataHandlerMixedSamplingWithFil
time_deltas = np.round(self.time_delta(self.cutoff_period)).astype(int)
start, end = window, 1
res = []
- window_array = self.create_index_array(self.window_dim.range(start, end), squeeze_dim=self.target_dim)
+ window_array = self.create_index_array(self.window_dim, range(start, end), squeeze_dim=self.target_dim)
for delta, filter_name in zip(np.append(time_deltas, 1), data.coords["filter"]):
res_filter = []
data_filter = data.sel({"filter": filter_name})
@@ -212,7 +226,7 @@ class DataHandlerSeparationOfScalesSingleStation(DataHandlerMixedSamplingWithFil
res_filter.append(data_filter.shift({dim: -w * delta}))
res_filter = xr.concat(res_filter, dim=window_array).chunk()
res.append(res_filter)
- res = xr.concat(res, dim="filter")
+ res = xr.concat(res, dim="filter").compute()
return res
def estimate_filter_width(self):
diff --git a/mlair/data_handler/data_handler_single_station.py b/mlair/data_handler/data_handler_single_station.py
index a894c635282b5879d79426168eb96d64ff5fa2a2..e9db27a9ff88efa2cc800723ac99279ec66d6cbb 100644
--- a/mlair/data_handler/data_handler_single_station.py
+++ b/mlair/data_handler/data_handler_single_station.py
@@ -5,9 +5,11 @@ __date__ = '2020-07-20'
import copy
import datetime as dt
+import dill
+import hashlib
import logging
import os
-from functools import reduce
+from functools import reduce, partial
from typing import Union, List, Iterable, Tuple, Dict, Optional
import numpy as np
@@ -45,6 +47,10 @@ class DataHandlerSingleStation(AbstractDataHandler):
DEFAULT_INTERPOLATION_LIMIT = 0
DEFAULT_INTERPOLATION_METHOD = "linear"
+ _hash = ["station", "statistics_per_var", "data_origin", "station_type", "network", "sampling", "target_dim",
+ "target_var", "time_dim", "iter_dim", "window_dim", "window_history_size", "window_history_offset",
+ "window_lead_time", "interpolation_limit", "interpolation_method"]
+
def __init__(self, station, data_path, statistics_per_var, station_type=DEFAULT_STATION_TYPE,
network=DEFAULT_NETWORK, sampling: Union[str, Tuple[str]] = DEFAULT_SAMPLING,
target_dim=DEFAULT_TARGET_DIM, target_var=DEFAULT_TARGET_VAR, time_dim=DEFAULT_TIME_DIM,
@@ -54,10 +60,16 @@ class DataHandlerSingleStation(AbstractDataHandler):
interpolation_limit: Union[int, Tuple[int]] = DEFAULT_INTERPOLATION_LIMIT,
interpolation_method: Union[str, Tuple[str]] = DEFAULT_INTERPOLATION_METHOD,
overwrite_local_data: bool = False, transformation=None, store_data_locally: bool = True,
- min_length: int = 0, start=None, end=None, variables=None, data_origin: Dict = None, **kwargs):
+ min_length: int = 0, start=None, end=None, variables=None, data_origin: Dict = None,
+ lazy_preprocessing: bool = False, **kwargs):
super().__init__()
self.station = helpers.to_list(station)
self.path = self.setup_data_path(data_path, sampling)
+ self.lazy = lazy_preprocessing
+ self.lazy_path = None
+ if self.lazy is True:
+ self.lazy_path = os.path.join(data_path, "lazy_data", self.__class__.__name__)
+ check_path_and_create(self.lazy_path)
self.statistics_per_var = statistics_per_var
self.data_origin = data_origin
self.do_transformation = transformation is not None
@@ -183,7 +195,17 @@ class DataHandlerSingleStation(AbstractDataHandler):
else:
raise NotImplementedError
- def f_apply(data, method, mean=None, std=None, min=None, max=None):
+ def f_apply(data, method, **kwargs):
+ for k, v in kwargs.items():
+ if not (isinstance(v, xr.DataArray) or v is None):
+ _, opts = statistics.min_max(data, dim)
+ helper = xr.ones_like(opts['min'])
+ kwargs[k] = helper * v
+ mean = kwargs.pop('mean', None)
+ std = kwargs.pop('std', None)
+ min = kwargs.pop('min', None)
+ max = kwargs.pop('max', None)
+
if method == "standardise":
return statistics.standardise_apply(data, mean, std), {"mean": mean, "std": std, "method": method}
elif method == "centre":
@@ -215,15 +237,48 @@ class DataHandlerSingleStation(AbstractDataHandler):
"""
Setup samples. This method prepares and creates samples X, and labels Y.
"""
+ if self.lazy is False:
+ self.make_input_target()
+ else:
+ self.load_lazy()
+ self.store_lazy()
+ if self.do_transformation is True:
+ self.call_transform()
+ self.make_samples()
+
+ def store_lazy(self):
+ hash = self._get_hash()
+ filename = os.path.join(self.lazy_path, hash + ".pickle")
+ if not os.path.exists(filename):
+ dill.dump(self._create_lazy_data(), file=open(filename, "wb"))
+
+ def _create_lazy_data(self):
+ return [self._data, self.meta, self.input_data, self.target_data]
+
+ def load_lazy(self):
+ hash = self._get_hash()
+ filename = os.path.join(self.lazy_path, hash + ".pickle")
+ try:
+ with open(filename, "rb") as pickle_file:
+ lazy_data = dill.load(pickle_file)
+ self._extract_lazy(lazy_data)
+ logging.debug(f"{self.station[0]}: used lazy data")
+ except FileNotFoundError:
+ logging.debug(f"{self.station[0]}: could not use lazy data")
+ self.make_input_target()
+
+ def _extract_lazy(self, lazy_data):
+ _data, self.meta, _input_data, _target_data = lazy_data
+ f_prep = partial(self._slice_prep, start=self.start, end=self.end)
+ self._data, self.input_data, self.target_data = list(map(f_prep, [_data, _input_data, _target_data]))
+
+ def make_input_target(self):
data, self.meta = self.load_data(self.path, self.station, self.statistics_per_var, self.sampling,
self.station_type, self.network, self.store_data_locally, self.data_origin,
self.start, self.end)
self._data = self.interpolate(data, dim=self.time_dim, method=self.interpolation_method,
limit=self.interpolation_limit)
self.set_inputs_and_targets()
- if self.do_transformation is True:
- self.call_transform()
- self.make_samples()
def set_inputs_and_targets(self):
inputs = self._data.sel({self.target_dim: helpers.to_list(self.variables)})
@@ -551,8 +606,7 @@ class DataHandlerSingleStation(AbstractDataHandler):
"""
return data.loc[{coord: slice(str(start), str(end))}]
- @staticmethod
- def setup_transformation(transformation: Union[None, dict, Tuple]) -> Tuple[Optional[dict], Optional[dict]]:
+ def setup_transformation(self, transformation: Union[None, dict, Tuple]) -> Tuple[Optional[dict], Optional[dict]]:
"""
Set up transformation by extracting all relevant information.
@@ -658,6 +712,13 @@ class DataHandlerSingleStation(AbstractDataHandler):
return self.transform(data, dim=dim, opts=self._transformation[pos], inverse=inverse,
transformation_dim=self.target_dim)
+ def _hash_list(self):
+ return sorted(list(set(self._hash)))
+
+ def _get_hash(self):
+ hash = "".join([str(self.__getattribute__(e)) for e in self._hash_list()]).encode()
+ return hashlib.md5(hash).hexdigest()
+
if __name__ == "__main__":
# dp = AbstractDataPrep('data/', 'dummy', 'DEBW107', ['o3', 'temp'], statistics_per_var={'o3': 'dma8eu', 'temp': 'maximum'})
diff --git a/mlair/data_handler/default_data_handler.py b/mlair/data_handler/default_data_handler.py
index 704b07cfb83baaa0539f134058d6096adc5554a9..ff1f13aed4b8d829edc653c6e99b6cff82287476 100644
--- a/mlair/data_handler/default_data_handler.py
+++ b/mlair/data_handler/default_data_handler.py
@@ -8,6 +8,7 @@ import gc
import logging
import os
import pickle
+import dill
import shutil
from functools import reduce
from typing import Tuple, Union, List
@@ -86,7 +87,7 @@ class DefaultDataHandler(AbstractDataHandler):
data = {"X": self._X, "Y": self._Y, "X_extreme": self._X_extreme, "Y_extreme": self._Y_extreme}
data = self._force_dask_computation(data)
with open(self._save_file, "wb") as f:
- pickle.dump(data, f)
+ dill.dump(data, f)
logging.debug(f"save pickle data to {self._save_file}")
self._reset_data()
@@ -101,7 +102,7 @@ class DefaultDataHandler(AbstractDataHandler):
def _load(self):
try:
with open(self._save_file, "rb") as f:
- data = pickle.load(f)
+ data = dill.load(f)
logging.debug(f"load pickle data from {self._save_file}")
self._X, self._Y = data["X"], data["Y"]
self._X_extreme, self._Y_extreme = data["X_extreme"], data["Y_extreme"]
@@ -240,6 +241,8 @@ class DefaultDataHandler(AbstractDataHandler):
* standardise (default, if method is not given)
* centre
+ * min_max
+ * log
### mean and std estimation
@@ -255,14 +258,16 @@ class DefaultDataHandler(AbstractDataHandler):
If mean and std are not None, the default data handler expects this parameters to match the data and applies
this values to the data. Make sure that all dimensions and/or coordinates are in agreement.
+
+ ### min and max given
+ If min and max are not None, the default data handler expects this parameters to match the data and applies
+ this values to the data. Make sure that all dimensions and/or coordinates are in agreement.
"""
sp_keys = {k: copy.deepcopy(kwargs[k]) for k in cls._requirements if k in kwargs}
- transformation_dict = sp_keys.get("transformation", None)
- if transformation_dict is None:
+ if "transformation" not in sp_keys.keys():
return
- if isinstance(transformation_dict, dict): # tuple for (input, target) transformation
- transformation_dict = copy.deepcopy(transformation_dict), copy.deepcopy(transformation_dict)
+ transformation_dict = ({}, {})
def _inner():
"""Inner method that is performed in both serial and parallel approach."""
@@ -272,7 +277,9 @@ class DefaultDataHandler(AbstractDataHandler):
if var not in transformation_dict[i].keys():
transformation_dict[i][var] = {}
opts = transformation[var]
- assert transformation_dict[i][var].get("method", opts["method"]) == opts["method"]
+ if not transformation_dict[i][var].get("method", opts["method"]) == opts["method"]:
+ # data handlers with filters are allowed to change transformation method to standardise
+ assert hasattr(dh, "filter_dim") and opts["method"] == "standardise"
transformation_dict[i][var]["method"] = opts["method"]
for k in ["mean", "std", "min", "max"]:
old = transformation_dict[i][var].get(k, None)
diff --git a/mlair/data_handler/iterator.py b/mlair/data_handler/iterator.py
index 30c45417a64e949b0c0535a96a20c933641fdcbb..564bf3bfd6e4f5b814c9d090733cfbfbf26a850b 100644
--- a/mlair/data_handler/iterator.py
+++ b/mlair/data_handler/iterator.py
@@ -9,6 +9,7 @@ import math
import os
import shutil
import pickle
+import dill
from typing import Tuple, List
@@ -109,7 +110,7 @@ class KerasIterator(keras.utils.Sequence):
"""Load pickle data from disk."""
file = self._path % index
with open(file, "rb") as f:
- data = pickle.load(f)
+ data = dill.load(f)
return data["X"], data["Y"]
@staticmethod
@@ -167,7 +168,7 @@ class KerasIterator(keras.utils.Sequence):
data = {"X": X, "Y": Y}
file = self._path % index
with open(file, "wb") as f:
- pickle.dump(data, f)
+ dill.dump(data, f)
def _get_number_of_mini_batches(self, number_of_samples: int) -> int:
"""Return number of mini batches as the floored ration of number of samples to batch size."""
diff --git a/mlair/helpers/helpers.py b/mlair/helpers/helpers.py
index ee727ef59ff35334be0a52a4d78dbae814d6c205..b57b733b08c4635a16d7fd18e99538a991521fd8 100644
--- a/mlair/helpers/helpers.py
+++ b/mlair/helpers/helpers.py
@@ -103,7 +103,7 @@ def remove_items(obj: Union[List, Dict], items: Any):
raise TypeError(f"{inspect.stack()[0][3]} does not support type {type(obj)}.")
-def select_from_dict(dict_obj: dict, sel_list: Any):
+def select_from_dict(dict_obj: dict, sel_list: Any, remove_none=False):
"""
Extract all key values pairs whose key is contained in the sel_list.
@@ -113,6 +113,7 @@ def select_from_dict(dict_obj: dict, sel_list: Any):
sel_list = to_list(sel_list)
assert isinstance(dict_obj, dict)
sel_dict = {k: v for k, v in dict_obj.items() if k in sel_list}
+ sel_dict = sel_dict if not remove_none else {k: v for k, v in sel_dict.items() if v is not None}
return sel_dict
diff --git a/mlair/helpers/statistics.py b/mlair/helpers/statistics.py
index 3631597aedb90b3411163a42490e9c023bad706a..3e99357c36d556f093701325964500bf8d46c698 100644
--- a/mlair/helpers/statistics.py
+++ b/mlair/helpers/statistics.py
@@ -11,8 +11,10 @@ import pandas as pd
from typing import Union, Tuple, Dict, List
from matplotlib import pyplot as plt
import itertools
+import gc
+import warnings
-from mlair.helpers import to_list
+from mlair.helpers import to_list, TimeTracking, TimeTrackingWrapper
Data = Union[xr.DataArray, pd.DataFrame]
@@ -438,7 +440,7 @@ class SkillScores:
"""Calculate CASE IV."""
AI, BI, CI, data, suffix = self.skill_score_pre_calculations(internal_data, observation_name, forecast_name)
monthly_mean_external = self.create_monthly_mean_from_daily_data(external_data, index=data.index)
- data = xr.concat([data, monthly_mean_external], dim="type")
+ data = xr.concat([data, monthly_mean_external], dim="type").dropna(dim="index")
mean, sigma = suffix["mean"], suffix["sigma"]
mean_external = monthly_mean_external.mean()
sigma_external = np.sqrt(monthly_mean_external.var())
@@ -608,6 +610,48 @@ class KolmogorovZurbenkoFilterMovingWindow(KolmogorovZurbenkoBaseClass):
else:
return None
+ @TimeTrackingWrapper
+ def kz_filter_new(self, df, wl, itr):
+ """
+ It passes the low frequency time series.
+
+ If filter method is from mean, max, min this method will call construct and rechunk before the actual
+ calculation to improve performance. If filter method is either median or percentile this approach is not
+ applicable and depending on the data and window size, this method can become slow.
+
+ Args:
+ wl(int): a window length
+ itr(int): a number of iteration
+ """
+ warnings.filterwarnings("ignore")
+ df_itr = df.__deepcopy__()
+ try:
+ kwargs = {"min_periods": int(0.7 * wl),
+ "center": True,
+ self.filter_dim: wl}
+ for i in np.arange(0, itr):
+ print(i)
+ rolling = df_itr.chunk().rolling(**kwargs)
+ if self.method not in ["percentile", "median"]:
+ rolling = rolling.construct("construct").chunk("auto")
+ if self.method == "median":
+ df_mv_avg_tmp = rolling.median()
+ elif self.method == "percentile":
+ df_mv_avg_tmp = rolling.quantile(self.percentile)
+ elif self.method == "max":
+ df_mv_avg_tmp = rolling.max("construct")
+ elif self.method == "min":
+ df_mv_avg_tmp = rolling.min("construct")
+ else:
+ df_mv_avg_tmp = rolling.mean("construct")
+ df_itr = df_mv_avg_tmp.compute()
+ del df_mv_avg_tmp, rolling
+ gc.collect()
+ return df_itr
+ except ValueError:
+ raise ValueError
+
+ @TimeTrackingWrapper
def kz_filter(self, df, wl, itr):
"""
It passes the low frequency time series.
@@ -616,15 +660,18 @@ class KolmogorovZurbenkoFilterMovingWindow(KolmogorovZurbenkoBaseClass):
wl(int): a window length
itr(int): a number of iteration
"""
+ import warnings
+ warnings.filterwarnings("ignore")
df_itr = df.__deepcopy__()
try:
- kwargs = {"min_periods": 1,
+ kwargs = {"min_periods": int(0.7 * wl),
"center": True,
self.filter_dim: wl}
iter_vars = df_itr.coords["variables"].values
for var in iter_vars:
- df_itr_var = df_itr.sel(variables=[var]).chunk()
+ df_itr_var = df_itr.sel(variables=[var])
for _ in np.arange(0, itr):
+ df_itr_var = df_itr_var.chunk()
rolling = df_itr_var.rolling(**kwargs)
if self.method == "median":
df_mv_avg_tmp = rolling.median()
@@ -637,7 +684,7 @@ class KolmogorovZurbenkoFilterMovingWindow(KolmogorovZurbenkoBaseClass):
else:
df_mv_avg_tmp = rolling.mean()
df_itr_var = df_mv_avg_tmp.compute()
- df_itr = df_itr.drop_sel(variables=var).combine_first(df_itr_var)
+ df_itr.loc[{"variables": [var]}] = df_itr_var
return df_itr
except ValueError:
raise ValueError
diff --git a/mlair/model_modules/abstract_model_class.py b/mlair/model_modules/abstract_model_class.py
index 894ff7ac4e787a8b31f75ff932f60bec8c561094..989f4578f78e6566dfca5a63f671ced8120491d8 100644
--- a/mlair/model_modules/abstract_model_class.py
+++ b/mlair/model_modules/abstract_model_class.py
@@ -82,7 +82,7 @@ class AbstractModelClass(ABC):
self.__custom_objects = value
@property
- def compile_options(self) -> Callable:
+ def compile_options(self) -> Dict:
"""
The compile options property allows the user to use all keras.compile() arguments. They can ether be passed as
dictionary (1), as attribute, without setting compile_options (2) or as mixture (partly defined as instance
@@ -116,7 +116,7 @@ class AbstractModelClass(ABC):
def set_compile_options(self):
self.optimizer = keras.optimizers.SGD()
self.loss = keras.losses.mean_squared_error
- self.compile_options = {"optimizer" = keras.optimizers.Adam(), "metrics": ["mse", "mae"]}
+ self.compile_options = {"optimizer": keras.optimizers.Adam(), "metrics": ["mse", "mae"]}
Note:
* As long as the attribute and the dict value have exactly the same values, the setter method will not raise
diff --git a/mlair/model_modules/fully_connected_networks.py b/mlair/model_modules/fully_connected_networks.py
index dbcd3a9f41ca1b9a7435be95b93eb40c2b37c5a0..9fb08cdf6efacab12c2828ed221966586bce1d08 100644
--- a/mlair/model_modules/fully_connected_networks.py
+++ b/mlair/model_modules/fully_connected_networks.py
@@ -5,57 +5,11 @@ from functools import reduce, partial
from mlair.model_modules import AbstractModelClass
from mlair.helpers import select_from_dict
+from mlair.model_modules.loss import var_loss, custom_loss
import keras
-class FCN_64_32_16(AbstractModelClass):
- """
- A customised model 4 Dense layers (64, 32, 16, window_lead_time), where the last layer is the output layer depending
- on the window_lead_time parameter.
- """
-
- def __init__(self, input_shape: list, output_shape: list):
- """
- Sets model and loss depending on the given arguments.
-
- :param input_shape: list of input shapes (expect len=1 with shape=(window_hist, station, variables))
- :param output_shape: list of output shapes (expect len=1 with shape=(window_forecast))
- """
-
- assert len(input_shape) == 1
- assert len(output_shape) == 1
- super().__init__(input_shape[0], output_shape[0])
-
- # settings
- self.activation = keras.layers.PReLU
-
- # apply to model
- self.set_model()
- self.set_compile_options()
- self.set_custom_objects(loss=self.compile_options['loss'])
-
- def set_model(self):
- """
- Build the model.
- """
- x_input = keras.layers.Input(shape=self._input_shape)
- x_in = keras.layers.Flatten()(x_input)
- x_in = keras.layers.Dense(64, name="Dense_64")(x_in)
- x_in = self.activation()(x_in)
- x_in = keras.layers.Dense(32, name="Dense_32")(x_in)
- x_in = self.activation()(x_in)
- x_in = keras.layers.Dense(16, name="Dense_16")(x_in)
- x_in = self.activation()(x_in)
- x_in = keras.layers.Dense(self._output_shape, name="Dense_output")(x_in)
- out_main = self.activation()(x_in)
- self.model = keras.Model(inputs=x_input, outputs=[out_main])
-
- def set_compile_options(self):
- self.optimizer = keras.optimizers.adam(lr=1e-2)
- self.compile_options = {"loss": [keras.losses.mean_squared_error], "metrics": ["mse", "mae"]}
-
-
class FCN(AbstractModelClass):
"""
A customisable fully connected network (64, 32, 16, window_lead_time), where the last layer is the output layer depending
@@ -64,12 +18,20 @@ class FCN(AbstractModelClass):
_activation = {"relu": keras.layers.ReLU, "tanh": partial(keras.layers.Activation, "tanh"),
"sigmoid": partial(keras.layers.Activation, "sigmoid"),
- "linear": partial(keras.layers.Activation, "linear")}
+ "linear": partial(keras.layers.Activation, "linear"),
+ "selu": partial(keras.layers.Activation, "selu"),
+ "prelu": partial(keras.layers.PReLU, alpha_initializer=keras.initializers.constant(value=0.25))}
+ _initializer = {"tanh": "glorot_uniform", "sigmoid": "glorot_uniform", "linear": "glorot_uniform",
+ "relu": keras.initializers.he_normal(), "selu": keras.initializers.lecun_normal(),
+ "prelu": keras.initializers.he_normal()}
_optimizer = {"adam": keras.optimizers.adam, "sgd": keras.optimizers.SGD}
- _requirements = ["lr", "beta_1", "beta_2", "epsilon", "decay", "amsgrad", "momentum", "nesterov"]
+ _regularizer = {"l1": keras.regularizers.l1, "l2": keras.regularizers.l2, "l1_l2": keras.regularizers.l1_l2}
+ _requirements = ["lr", "beta_1", "beta_2", "epsilon", "decay", "amsgrad", "momentum", "nesterov", "l1", "l2"]
+ _dropout = {"selu": keras.layers.AlphaDropout}
def __init__(self, input_shape: list, output_shape: list, activation="relu", activation_output="linear",
- optimizer="adam", n_layer=1, n_hidden=10, **kwargs):
+ optimizer="adam", n_layer=1, n_hidden=10, regularizer=None, dropout=None, layer_configuration=None,
+ **kwargs):
"""
Sets model and loss depending on the given arguments.
@@ -83,15 +45,20 @@ class FCN(AbstractModelClass):
# settings
self.activation = self._set_activation(activation)
+ self.activation_name = activation
self.activation_output = self._set_activation(activation_output)
+ self.activation_output_name = activation_output
self.optimizer = self._set_optimizer(optimizer, **kwargs)
- self.layer_configuration = (n_layer, n_hidden)
+ self.layer_configuration = (n_layer, n_hidden) if layer_configuration is None else layer_configuration
self._update_model_name()
+ self.kernel_initializer = self._initializer.get(activation, "glorot_uniform")
+ self.kernel_regularizer = self._set_regularizer(regularizer, **kwargs)
+ self.dropout, self.dropout_rate = self._set_dropout(activation, dropout)
# apply to model
self.set_model()
self.set_compile_options()
- # self.set_custom_objects(loss=self.compile_options['loss'])
+ self.set_custom_objects(loss=self.compile_options["loss"][0], var_loss=var_loss)
def _set_activation(self, activation):
try:
@@ -112,11 +79,37 @@ class FCN(AbstractModelClass):
except KeyError:
raise AttributeError(f"Given optimizer {optimizer} is not supported in this model class.")
+ def _set_regularizer(self, regularizer, **kwargs):
+ if regularizer is None or (isinstance(regularizer, str) and regularizer.lower() == "none"):
+ return None
+ try:
+ reg_name = regularizer.lower()
+ reg = self._regularizer.get(reg_name)
+ reg_kwargs = {}
+ if reg_name in ["l1", "l2"]:
+ reg_kwargs = select_from_dict(kwargs, reg_name, remove_none=True)
+ if reg_name in reg_kwargs:
+ reg_kwargs["l"] = reg_kwargs.pop(reg_name)
+ elif reg_name == "l1_l2":
+ reg_kwargs = select_from_dict(kwargs, ["l1", "l2"], remove_none=True)
+ return reg(**reg_kwargs)
+ except KeyError:
+ raise AttributeError(f"Given regularizer {regularizer} is not supported in this model class.")
+
+ def _set_dropout(self, activation, dropout_rate):
+ if dropout_rate is None:
+ return None, None
+ assert 0 <= dropout_rate < 1
+ return self._dropout.get(activation, keras.layers.Dropout), dropout_rate
+
def _update_model_name(self):
- n_layer, n_hidden = self.layer_configuration
n_input = str(reduce(lambda x, y: x * y, self._input_shape))
n_output = str(self._output_shape)
- self.model_name += "_".join(["", n_input, *[f"{n_hidden}" for _ in range(n_layer)], n_output])
+ if isinstance(self.layer_configuration, tuple) and len(self.layer_configuration) == 2:
+ n_layer, n_hidden = self.layer_configuration
+ self.model_name += "_".join(["", n_input, *[f"{n_hidden}" for _ in range(n_layer)], n_output])
+ else:
+ self.model_name += "_".join(["", n_input, *[f"{n}" for n in self.layer_configuration], n_output])
def set_model(self):
"""
@@ -124,13 +117,53 @@ class FCN(AbstractModelClass):
"""
x_input = keras.layers.Input(shape=self._input_shape)
x_in = keras.layers.Flatten()(x_input)
- n_layer, n_hidden = self.layer_configuration
- for layer in range(n_layer):
- x_in = keras.layers.Dense(n_hidden)(x_in)
- x_in = self.activation()(x_in)
+ if isinstance(self.layer_configuration, tuple) is True:
+ n_layer, n_hidden = self.layer_configuration
+ for layer in range(n_layer):
+ x_in = keras.layers.Dense(n_hidden, kernel_initializer=self.kernel_initializer,
+ kernel_regularizer=self.kernel_regularizer)(x_in)
+ x_in = self.activation(name=f"{self.activation_name}_{layer + 1}")(x_in)
+ if self.dropout is not None:
+ x_in = self.dropout(self.dropout_rate)(x_in)
+ else:
+ assert isinstance(self.layer_configuration, list) is True
+ for layer, n_hidden in enumerate(self.layer_configuration):
+ x_in = keras.layers.Dense(n_hidden, kernel_initializer=self.kernel_initializer,
+ kernel_regularizer=self.kernel_regularizer)(x_in)
+ x_in = self.activation(name=f"{self.activation_name}_{layer + 1}")(x_in)
+ if self.dropout is not None:
+ x_in = self.dropout(self.dropout_rate)(x_in)
x_in = keras.layers.Dense(self._output_shape)(x_in)
- out = self.activation_output()(x_in)
+ out = self.activation_output(name=f"{self.activation_output_name}_output")(x_in)
self.model = keras.Model(inputs=x_input, outputs=[out])
+ def set_compile_options(self):
+ self.compile_options = {"loss": [custom_loss([keras.losses.mean_squared_error, var_loss])],
+ "metrics": ["mse", "mae", var_loss]}
+
+
+class FCN_64_32_16(FCN):
+ """
+ A customised model 4 Dense layers (64, 32, 16, window_lead_time), where the last layer is the output layer depending
+ on the window_lead_time parameter.
+ """
+
+ _requirements = ["lr", "beta_1", "beta_2", "epsilon", "decay", "amsgrad"]
+
+ def __init__(self, input_shape: list, output_shape: list, **kwargs):
+ """
+ Sets model and loss depending on the given arguments.
+
+ :param input_shape: list of input shapes (expect len=1 with shape=(window_hist, station, variables))
+ :param output_shape: list of output shapes (expect len=1 with shape=(window_forecast))
+ """
+ lr = kwargs.pop("lr", 1e-2)
+ super().__init__(input_shape, output_shape, activation="prelu", activation_output="linear",
+ layer_configuration=[64, 32, 16], optimizer="adam", lr=lr, **kwargs)
+
def set_compile_options(self):
self.compile_options = {"loss": [keras.losses.mean_squared_error], "metrics": ["mse", "mae"]}
+
+ def _update_model_name(self):
+ self.model_name = "FCN"
+ super()._update_model_name()
diff --git a/mlair/model_modules/loss.py b/mlair/model_modules/loss.py
index bcb85282d0fa15f18ebd65a89e4020c2a0170224..ba871e983ecfa1e91676d53b834ebd622c00fe49 100644
--- a/mlair/model_modules/loss.py
+++ b/mlair/model_modules/loss.py
@@ -20,3 +20,21 @@ def l_p_loss(power: int) -> Callable:
return K.mean(K.pow(K.abs(y_pred - y_true), power), axis=-1)
return loss
+
+
+def var_loss(y_true, y_pred) -> Callable:
+ return K.mean(K.square(K.var(y_true) - K.var(y_pred)))
+
+
+def custom_loss(loss_list, loss_weights=None) -> Callable:
+ n = len(loss_list)
+ if loss_weights is None:
+ loss_weights = [1. / n for _ in range(n)]
+ else:
+ assert len(loss_weights) == n
+ loss_weights = [w / sum(loss_weights) for w in loss_weights]
+
+ def loss(y_true, y_pred):
+ return sum([loss_weights[i] * loss_list[i](y_true, y_pred) for i in range(n)])
+
+ return loss
diff --git a/mlair/run_modules/post_processing.py b/mlair/run_modules/post_processing.py
index 3b9b563426a80816f7cf1ea9e114a8395d9fbba0..73aebb008ebf1f61eb2878293fc160cf549d19cb 100644
--- a/mlair/run_modules/post_processing.py
+++ b/mlair/run_modules/post_processing.py
@@ -306,7 +306,7 @@ class PostProcessing(RunEnvironment):
try:
if ("filter" in self.test_data[0].get_X(as_numpy=False)[0].coords) and (
"PlotSeparationOfScales" in plot_list):
- filter_dim = self.data_store.get("filter_dim", None)
+ filter_dim = self.data_store.get_default("filter_dim", None)
PlotSeparationOfScales(self.test_data, plot_folder=self.plot_path, time_dim=time_dim,
window_dim=window_dim, target_dim=target_dim, **{"filter_dim": filter_dim})
except Exception as e:
diff --git a/requirements.txt b/requirements.txt
index b0a6e7f59896fd0edf08977ee553c803f6c2e960..85655e237f8e10e98f77c379be6acd0a7bb65d46 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -2,6 +2,7 @@ absl-py==0.11.0
appdirs==1.4.4
astor==0.8.1
attrs==20.3.0
+bottleneck==1.3.2
cached-property==1.5.2
certifi==2020.12.5
cftime==1.4.1
@@ -9,6 +10,7 @@ chardet==4.0.0
coverage==5.4
cycler==0.10.0
dask==2021.2.0
+dill==0.3.3
fsspec==0.8.5
gast==0.4.0
grpcio==1.35.0
diff --git a/requirements_gpu.txt b/requirements_gpu.txt
index 35fe0d5ee2a03f01737bc185d2a5bbaf26383806..cc189496bdf4e1e1ee86902a1953c2058d58c8e4 100644
--- a/requirements_gpu.txt
+++ b/requirements_gpu.txt
@@ -2,6 +2,7 @@ absl-py==0.11.0
appdirs==1.4.4
astor==0.8.1
attrs==20.3.0
+bottleneck==1.3.2
cached-property==1.5.2
certifi==2020.12.5
cftime==1.4.1
@@ -9,6 +10,7 @@ chardet==4.0.0
coverage==5.4
cycler==0.10.0
dask==2021.2.0
+dill==0.3.3
fsspec==0.8.5
gast==0.4.0
grpcio==1.35.0
diff --git a/test/test_data_handler/test_data_handler_mixed_sampling.py b/test/test_data_handler/test_data_handler_mixed_sampling.py
index d2f9ce00224a61815c89e44b7c37a667d239b2f5..2a6553b7f495bb4eb8aeddf7c39f2f2517edc967 100644
--- a/test/test_data_handler/test_data_handler_mixed_sampling.py
+++ b/test/test_data_handler/test_data_handler_mixed_sampling.py
@@ -37,7 +37,7 @@ class TestDataHandlerMixedSamplingSingleStation:
req = object.__new__(DataHandlerSingleStation)
assert sorted(obj._requirements) == sorted(remove_items(req.requirements(), "station"))
- @mock.patch("mlair.data_handler.data_handler_mixed_sampling.DataHandlerMixedSamplingSingleStation.setup_samples")
+ @mock.patch("mlair.data_handler.data_handler_single_station.DataHandlerSingleStation.setup_samples")
def test_init(self, mock_super_init):
obj = DataHandlerMixedSamplingSingleStation("first_arg", "second", {}, test=23, sampling="hourly",
interpolation_limit=(1, 10))
diff --git a/test/test_helpers/test_helpers.py b/test/test_helpers/test_helpers.py
index f2e2b341afa424ce351c0253f41c75e362b77eba..91f2278ae7668b623f8d2434ebac7e959dc9c805 100644
--- a/test/test_helpers/test_helpers.py
+++ b/test/test_helpers/test_helpers.py
@@ -175,7 +175,7 @@ class TestSelectFromDict:
@pytest.fixture
def dictionary(self):
- return {"a": 1, "b": 23, "c": "last"}
+ return {"a": 1, "b": 23, "c": "last", "e": None}
def test_select(self, dictionary):
assert select_from_dict(dictionary, "c") == {"c": "last"}
@@ -186,6 +186,10 @@ class TestSelectFromDict:
with pytest.raises(AssertionError):
select_from_dict(["we"], "now")
+ def test_select_remove_none(self, dictionary):
+ assert select_from_dict(dictionary, ["a", "e"]) == {"a": 1, "e": None}
+ assert select_from_dict(dictionary, ["a", "e"], remove_none=True) == {"a": 1}
+
class TestRemoveItems:
diff --git a/test/test_model_modules/test_loss.py b/test/test_model_modules/test_loss.py
index e54e0b00de4a71d241f30e0b6b0c1a2e8fa1a19c..c993830c5290c9beeec392dfd806354ca02eb490 100644
--- a/test/test_model_modules/test_loss.py
+++ b/test/test_model_modules/test_loss.py
@@ -1,10 +1,12 @@
import keras
import numpy as np
-from mlair.model_modules.loss import l_p_loss
+from mlair.model_modules.loss import l_p_loss, var_loss, custom_loss
+import pytest
-class TestLoss:
+
+class TestLPLoss:
def test_l_p_loss(self):
model = keras.Sequential()
@@ -14,4 +16,42 @@ class TestLoss:
assert hist.history['loss'][0] == 1.25
model.compile(optimizer=keras.optimizers.Adam(), loss=l_p_loss(3))
hist = model.fit(np.array([1, 0, -2, 0.5]), np.array([1, 1, 0, 0.5]), epochs=1)
- assert hist.history['loss'][0] == 2.25
\ No newline at end of file
+ assert hist.history['loss'][0] == 2.25
+
+
+class TestVarLoss:
+
+ def test_var_loss(self):
+ model = keras.Sequential()
+ model.add(keras.layers.Lambda(lambda x: x, input_shape=(None,)))
+ model.compile(optimizer=keras.optimizers.Adam(), loss=var_loss)
+ hist = model.fit(np.array([1, 0, 2, 0.5]), np.array([1, 1, 0, 0.5]), epochs=1)
+ assert hist.history['loss'][0] == 0.140625
+
+
+class TestCustomLoss:
+
+ def test_custom_loss_no_weights(self):
+ cust_loss = custom_loss([l_p_loss(2), var_loss])
+ model = keras.Sequential()
+ model.add(keras.layers.Lambda(lambda x: x, input_shape=(None,)))
+ model.compile(optimizer=keras.optimizers.Adam(), loss=cust_loss)
+ hist = model.fit(np.array([1, 0, 2, 0.5]), np.array([1, 1, 0, 0.5]), epochs=1)
+ assert hist.history['loss'][0] == (0.5 * 0.140625 + 0.5 * 1.25)
+
+ @pytest.mark.parametrize("weights", [[0.3, 0.7], [0.5, 0.5], [1, 1], [4, 1]])
+ def test_custom_loss_with_weights(self, weights):
+ cust_loss = custom_loss([l_p_loss(2), var_loss], weights)
+ model = keras.Sequential()
+ model.add(keras.layers.Lambda(lambda x: x, input_shape=(None,)))
+ model.compile(optimizer=keras.optimizers.Adam(), loss=cust_loss)
+ hist = model.fit(np.array([1, 0, 2, 0.5]), np.array([1, 1, 0, 0.5]), epochs=1)
+ weights_adjusted = list(map(lambda x: x / sum(weights), weights))
+ expected = (weights_adjusted[0] * 1.25 + weights_adjusted[1] * 0.140625)
+ assert np.testing.assert_almost_equal(hist.history['loss'][0], expected, decimal=6) is None
+
+ def test_custom_loss_invalid_weights(self):
+ with pytest.raises(AssertionError):
+ custom_loss([l_p_loss(2), var_loss], [0.3])
+ with pytest.raises(AssertionError):
+ custom_loss([l_p_loss(2), var_loss], [0.4, 3, 1])