diff --git a/mlair/configuration/.gitignore b/mlair/configuration/.gitignore
index 8e2358dc56797578fe0de020aa827b1fef8663bf..91eccc695f4ea58374a14a1ba0272f98f210c203 100644
--- a/mlair/configuration/.gitignore
+++ b/mlair/configuration/.gitignore
@@ -1 +1,2 @@
-join_settings.py
\ No newline at end of file
+join_settings.py
+join_rest
\ No newline at end of file
diff --git a/mlair/configuration/defaults.py b/mlair/configuration/defaults.py
index d191af2edd8a6fe2c1093b3f1c3f5d419cc42b76..51d4beafbbc0b346331db80567946c3acc702b8e 100644
--- a/mlair/configuration/defaults.py
+++ b/mlair/configuration/defaults.py
@@ -1,6 +1,7 @@
__author__ = "Lukas Leufen"
__date__ = '2020-06-25'
+from mlair.helpers.statistics import TransformationClass
DEFAULT_STATIONS = ['DEBW107', 'DEBY081', 'DEBW013', 'DEBW076', 'DEBW087']
DEFAULT_VAR_ALL_DICT = {'o3': 'dma8eu', 'relhum': 'average_values', 'temp': 'maximum', 'u': 'average_values',
@@ -13,8 +14,7 @@ DEFAULT_START = "1997-01-01"
DEFAULT_END = "2017-12-31"
DEFAULT_WINDOW_HISTORY_SIZE = 13
DEFAULT_OVERWRITE_LOCAL_DATA = False
-# DEFAULT_TRANSFORMATION = {"scope": "data", "method": "standardise", "mean": "estimate"}
-DEFAULT_TRANSFORMATION = {"scope": "data", "method": "standardise"}
+DEFAULT_TRANSFORMATION = TransformationClass(inputs_method="standardise", targets_method="standardise")
DEFAULT_HPC_LOGIN_LIST = ["ju", "hdfmll"] # ju[wels} #hdfmll(ogin)
DEFAULT_HPC_HOST_LIST = ["jw", "hdfmlc"] # first part of node names for Juwels (jw[comp], hdfmlc(ompute).
DEFAULT_CREATE_NEW_MODEL = True
@@ -46,15 +46,11 @@ DEFAULT_USE_ALL_STATIONS_ON_ALL_DATA_SETS = True
DEFAULT_EVALUATE_BOOTSTRAPS = True
DEFAULT_CREATE_NEW_BOOTSTRAPS = False
DEFAULT_NUMBER_OF_BOOTSTRAPS = 20
-#DEFAULT_PLOT_LIST = ["PlotMonthlySummary", "PlotStationMap", "PlotClimatologicalSkillScore", "PlotTimeSeries",
-# "PlotCompetitiveSkillScore", "PlotBootstrapSkillScore", "PlotConditionalQuantiles",
-# "PlotAvailability"]
-DEFAULT_PLOT_LIST = ["PlotMonthlySummary", "PlotStationMap", "PlotClimatologicalSkillScore",
+DEFAULT_PLOT_LIST = ["PlotMonthlySummary", "PlotStationMap", "PlotClimatologicalSkillScore", "PlotTimeSeries",
"PlotCompetitiveSkillScore", "PlotBootstrapSkillScore", "PlotConditionalQuantiles",
"PlotAvailability"]
-
def get_defaults():
"""Return all default parameters set in defaults.py"""
return {key: value for key, value in globals().items() if key.startswith('DEFAULT')}
diff --git a/mlair/data_handler/__init__.py b/mlair/data_handler/__init__.py
index 01d660031bbbdda08eba80044a08fcb034d8171b..495b6e7c8604a839a084a2b78a54563c13eb06e6 100644
--- a/mlair/data_handler/__init__.py
+++ b/mlair/data_handler/__init__.py
@@ -13,4 +13,4 @@ from .bootstraps import BootStraps
from .iterator import KerasIterator, DataCollection
from .default_data_handler import DefaultDataHandler
from .abstract_data_handler import AbstractDataHandler
-from .data_preparation_neighbors import DataHandlerNeighbors
+from .data_handler_neighbors import DataHandlerNeighbors
diff --git a/mlair/data_handler/abstract_data_handler.py b/mlair/data_handler/abstract_data_handler.py
index 04b3d4651347759130da15a05056f6ace3d0fc1f..26ccf69c85e999c540e656a2ceac5737390a579e 100644
--- a/mlair/data_handler/abstract_data_handler.py
+++ b/mlair/data_handler/abstract_data_handler.py
@@ -27,7 +27,10 @@ class AbstractDataHandler:
@classmethod
def own_args(cls, *args):
- return remove_items(inspect.getfullargspec(cls).args, ["self"] + list(args))
+ """Return all arguments (including kwonlyargs)."""
+ arg_spec = inspect.getfullargspec(cls)
+ list_of_args = arg_spec.args + arg_spec.kwonlyargs
+ return remove_items(list_of_args, ["self"] + list(args))
@classmethod
def transformation(cls, *args, **kwargs):
diff --git a/mlair/data_handler/advanced_data_handler.py b/mlair/data_handler/advanced_data_handler.py
index c2d210bffdb598b23c025f60b903ddef84e4509d..f04748e82f11116b265796afba7f401c1cad9342 100644
--- a/mlair/data_handler/advanced_data_handler.py
+++ b/mlair/data_handler/advanced_data_handler.py
@@ -10,15 +10,18 @@ import datetime as dt
from mlair.data_handler import AbstractDataHandler
-from typing import Union, List
+from typing import Union, List, Tuple, Dict
+import logging
+from functools import reduce
+from mlair.helpers.join import EmptyQueryResult
+from mlair.helpers import TimeTracking
number = Union[float, int]
num_or_list = Union[number, List[number]]
def run_data_prep():
-
- from .data_preparation_neighbors import DataHandlerNeighbors
+ from .data_handler_neighbors import DataHandlerNeighbors
data = DummyDataHandler("main_class")
data.get_X()
data.get_Y()
@@ -33,8 +36,7 @@ def run_data_prep():
def create_data_prep():
-
- from .data_preparation_neighbors import DataHandlerNeighbors
+ from .data_handler_neighbors import DataHandlerNeighbors
path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "testdata")
station_type = None
network = 'UBA'
@@ -98,7 +100,7 @@ class DummyDataHandler(AbstractDataHandler):
if __name__ == "__main__":
- from mlair.data_handler.station_preparation import DataHandlerSingleStation
+ from mlair.data_handler.data_handler_single_station import DataHandlerSingleStation
from mlair.data_handler.iterator import KerasIterator, DataCollection
data_prep = create_data_prep()
data_collection = DataCollection(data_prep)
diff --git a/mlair/data_handler/data_handler_kz_filter.py b/mlair/data_handler/data_handler_kz_filter.py
new file mode 100644
index 0000000000000000000000000000000000000000..ce96a8f5c039b5a232aa56765209927dd4019168
--- /dev/null
+++ b/mlair/data_handler/data_handler_kz_filter.py
@@ -0,0 +1,90 @@
+"""Data Handler using kz-filtered data."""
+
+__author__ = 'Lukas Leufen'
+__date__ = '2020-08-26'
+
+import inspect
+import numpy as np
+import pandas as pd
+import xarray as xr
+from typing import List, Union
+
+from mlair.data_handler.data_handler_single_station import DataHandlerSingleStation
+from mlair.data_handler import DefaultDataHandler
+from mlair.helpers import remove_items, to_list, TimeTrackingWrapper
+from mlair.helpers.statistics import KolmogorovZurbenkoFilterMovingWindow as KZFilter
+
+# define a more general date type for type hinting
+str_or_list = Union[str, List[str]]
+
+
+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"])
+
+ def __init__(self, *args, kz_filter_length, kz_filter_iter, **kwargs):
+ assert kwargs.get("sampling") == "hourly" # This data handler requires hourly data resolution
+ kz_filter_length = to_list(kz_filter_length)
+ kz_filter_iter = to_list(kz_filter_iter)
+ # self.original_data = None # ToDo: implement here something to store unfiltered data
+ self.kz_filter_length = kz_filter_length
+ self.kz_filter_iter = kz_filter_iter
+ self.cutoff_period = None
+ self.cutoff_period_days = None
+ super().__init__(*args, **kwargs)
+
+ def setup_samples(self):
+ """
+ Setup samples. This method prepares and creates samples X, and labels Y.
+ """
+ self.load_data()
+ self.interpolate(dim=self.time_dim, method=self.interpolation_method, limit=self.interpolation_limit)
+ self.set_inputs_and_targets()
+ self.apply_kz_filter()
+ # this is just a code snippet to check the results of the kz filter
+ # import matplotlib
+ # matplotlib.use("TkAgg")
+ # import matplotlib.pyplot as plt
+ # self.input_data.data.sel(filter="74d", variables="temp", Stations="DEBW107").plot()
+ # self.input_data.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):
+ """Apply kolmogorov zurbenko filter only on inputs."""
+ kz = KZFilter(self.input_data.data, wl=self.kz_filter_length, itr=self.kz_filter_iter, filter_dim="datetime")
+ filtered_data: List[xr.DataArray] = kz.run()
+ self.cutoff_period = kz.period_null()
+ self.cutoff_period_days = kz.period_null_days()
+ self.input_data.data = xr.concat(filtered_data, pd.Index(self.create_filter_index(), name="filter"))
+
+ def create_filter_index(self) -> pd.Index:
+ """
+ Round cut off periods in days and append 'res' for residuum index.
+
+ Round small numbers (<10) to single decimal, and higher numbers to int. Transform as list of str and append
+ 'res' for residuum index.
+ """
+ index = np.round(self.cutoff_period_days, 1)
+ f = lambda x: int(np.round(x)) if x >= 10 else np.round(x, 1)
+ index = list(map(f, index.tolist()))
+ index = list(map(lambda x: str(x) + "d", index)) + ["res"]
+ return pd.Index(index, name="filter")
+
+ def get_transposed_history(self) -> xr.DataArray:
+ """Return history.
+
+ :return: history with dimensions datetime, window, Stations, variables.
+ """
+ return self.history.transpose("datetime", "window", "Stations", "variables", "filter").copy()
+
+
+class DataHandlerKzFilter(DefaultDataHandler):
+ """Data handler using kz filtered data."""
+
+ data_handler = DataHandlerKzFilterSingleStation
+ data_handler_transformation = DataHandlerKzFilterSingleStation
+ _requirements = data_handler.requirements()
diff --git a/mlair/data_handler/data_preparation_neighbors.py b/mlair/data_handler/data_handler_neighbors.py
similarity index 85%
rename from mlair/data_handler/data_preparation_neighbors.py
rename to mlair/data_handler/data_handler_neighbors.py
index 1482bb9fe20afcc2b92d2b91ae523a6dca19c54d..a004e659969232a080d49eb6905007d353bbe99c 100644
--- a/mlair/data_handler/data_preparation_neighbors.py
+++ b/mlair/data_handler/data_handler_neighbors.py
@@ -4,9 +4,9 @@ __date__ = '2020-07-17'
from mlair.helpers import to_list
-from mlair.data_handler.station_preparation import DataHandlerSingleStation
from mlair.data_handler import DefaultDataHandler
import os
+import copy
from typing import Union, List
@@ -15,6 +15,7 @@ num_or_list = Union[number, List[number]]
class DataHandlerNeighbors(DefaultDataHandler):
+ """Data handler including neighboring stations."""
def __init__(self, id_class, data_path, neighbors=None, min_length=0,
extreme_values: num_or_list = None, extremes_on_right_tail_only: bool = False):
@@ -24,14 +25,14 @@ class DataHandlerNeighbors(DefaultDataHandler):
@classmethod
def build(cls, station, **kwargs):
- sp_keys = {k: kwargs[k] for k in cls._requirements if k in kwargs}
- sp = DataHandlerSingleStation(station, **sp_keys)
+ sp_keys = {k: copy.deepcopy(kwargs[k]) for k in cls._requirements if k in kwargs}
+ sp = cls.data_handler(station, **sp_keys)
n_list = []
for neighbor in kwargs.get("neighbors", []):
- n_list.append(DataHandlerSingleStation(neighbor, **sp_keys))
+ n_list.append(cls.data_handler(neighbor, **sp_keys))
else:
kwargs["neighbors"] = n_list if len(n_list) > 0 else None
- dp_args = {k: kwargs[k] for k in cls.own_args("id_class") if k in kwargs}
+ dp_args = {k: copy.deepcopy(kwargs[k]) for k in cls.own_args("id_class") if k in kwargs}
return cls(sp, **dp_args)
def _create_collection(self):
diff --git a/mlair/data_handler/station_preparation.py b/mlair/data_handler/data_handler_single_station.py
similarity index 74%
rename from mlair/data_handler/station_preparation.py
rename to mlair/data_handler/data_handler_single_station.py
index f3428e91bae3dc1d94a45dd7ff2bf931cff1fa54..460d1c100dadbc2aea5d43932e902cc080177b27 100644
--- a/mlair/data_handler/station_preparation.py
+++ b/mlair/data_handler/data_handler_single_station.py
@@ -3,6 +3,7 @@
__author__ = 'Lukas Leufen, Felix Kleinert'
__date__ = '2020-07-20'
+import copy
import datetime as dt
import logging
import os
@@ -15,7 +16,7 @@ import xarray as xr
from mlair.configuration import check_path_and_create
from mlair import helpers
-from mlair.helpers import join, statistics
+from mlair.helpers import join, statistics, TimeTrackingWrapper
from mlair.data_handler.abstract_data_handler import AbstractDataHandler
# define a more general date type for type hinting
@@ -53,7 +54,8 @@ class DataHandlerSingleStation(AbstractDataHandler):
self.station = helpers.to_list(station)
self.path = os.path.abspath(data_path)
self.statistics_per_var = statistics_per_var
- self.transformation = self.setup_transformation(transformation)
+ self.do_transformation = transformation is not None
+ self.input_data, self.target_data = self.setup_transformation(transformation)
self.station_type = station_type
self.network = network
@@ -74,20 +76,13 @@ class DataHandlerSingleStation(AbstractDataHandler):
self.end = end
# internal
- self.data = None
+ self._data: xr.DataArray = None # loaded raw data
self.meta = None
self.variables = list(statistics_per_var.keys()) if variables is None else variables
self.history = None
self.label = None
self.observation = None
- # internal for transformation
- self.mean = None
- self.std = None
- self.max = None
- self.min = None
- self._transform_method = None
-
# create samples
self.setup_samples()
@@ -100,7 +95,7 @@ class DataHandlerSingleStation(AbstractDataHandler):
@property
def shape(self):
- return self.data.shape, self.get_X().shape, self.get_Y().shape
+ return self._data.shape, self.get_X().shape, self.get_Y().shape
def __repr__(self):
return f"StationPrep(station={self.station}, data_path='{self.path}', " \
@@ -109,24 +104,7 @@ class DataHandlerSingleStation(AbstractDataHandler):
f"sampling='{self.sampling}', target_dim='{self.target_dim}', target_var='{self.target_var}', " \
f"time_dim='{self.time_dim}', window_history_size={self.window_history_size}, " \
f"window_lead_time={self.window_lead_time}, interpolation_limit={self.interpolation_limit}, " \
- f"interpolation_method='{self.interpolation_method}', overwrite_local_data={self.overwrite_local_data}, " \
- f"transformation={self._print_transformation_as_string})"
-
- @property
- def _print_transformation_as_string(self):
- str_name = ''
- if self.transformation is None:
- str_name = f'{None}'
- else:
- for k, v in self.transformation.items():
- if v is not None:
- try:
- v_pr = f"xr.DataArray.from_dict({v.to_dict()})"
- except AttributeError:
- v_pr = f"'{v}'"
- str_name += f"'{k}':{v_pr}, "
- str_name = f"{{{str_name}}}"
- return str_name
+ f"interpolation_method='{self.interpolation_method}', overwrite_local_data={self.overwrite_local_data})"
def get_transposed_history(self) -> xr.DataArray:
"""Return history.
@@ -153,29 +131,29 @@ class DataHandlerSingleStation(AbstractDataHandler):
return coords.rename(index={"station_lon": "lon", "station_lat": "lat"}).to_dict()[str(self)]
def call_transform(self, inverse=False):
- self.transform(dim=self.time_dim, method=self.transformation["method"],
- mean=self.transformation['mean'], std=self.transformation["std"],
- min_val=self.transformation["min"], max_val=self.transformation["max"],
- inverse=inverse
- )
-
- def set_transformation(self, transformation: dict):
- if self._transform_method is not None:
- self.call_transform(inverse=True)
- self.transformation = self.setup_transformation(transformation)
- self.call_transform()
- self.make_samples()
+ kwargs = helpers.remove_items(self.input_data.as_dict(), ["data"])
+ self.transform(self.input_data, dim=self.time_dim, inverse=inverse, **kwargs)
+ kwargs = helpers.remove_items(self.target_data.as_dict(), ["data"])
+ self.transform(self.target_data, dim=self.time_dim, inverse=inverse, **kwargs)
+ @TimeTrackingWrapper
def setup_samples(self):
"""
Setup samples. This method prepares and creates samples X, and labels Y.
"""
self.load_data()
self.interpolate(dim=self.time_dim, method=self.interpolation_method, limit=self.interpolation_limit)
- if self.transformation is not None:
+ 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)})
+ targets = self._data.sel({self.target_dim: self.target_var})
+ self.input_data.data = inputs
+ self.target_data.data = targets
+
def make_samples(self):
self.make_history_window(self.target_dim, self.window_history_size, self.time_dim)
self.make_labels(self.target_dim, self.target_var, self.time_dim, self.window_lead_time)
@@ -216,7 +194,7 @@ class DataHandlerSingleStation(AbstractDataHandler):
logging.debug("loading finished")
# create slices and check for negative concentration.
data = self._slice_prep(data)
- self.data = self.check_for_negative_concentrations(data)
+ self._data = self.check_for_negative_concentrations(data)
def download_data_from_join(self, file_name: str, meta_file: str) -> [xr.DataArray, pd.DataFrame]:
"""
@@ -283,10 +261,11 @@ class DataHandlerSingleStation(AbstractDataHandler):
data.loc[..., used_chem_vars] = data.loc[..., used_chem_vars].clip(min=minimum)
return data
- def shift(self, dim: str, window: int) -> xr.DataArray:
+ def shift(self, data: xr.DataArray, dim: str, window: int) -> xr.DataArray:
"""
Shift data multiple times to represent history (if window <= 0) or lead time (if window > 0).
+ :param data: data set to shift
:param dim: dimension along shift is applied
:param window: number of steps to shift (corresponds to the window length)
@@ -300,7 +279,7 @@ class DataHandlerSingleStation(AbstractDataHandler):
end = window + 1
res = []
for w in range(start, end):
- res.append(self.data.shift({dim: -w}))
+ res.append(data.shift({dim: -w}))
window_array = self.create_index_array('window', range(start, end), squeeze_dim=self.target_dim)
res = xr.concat(res, dim=window_array)
return res
@@ -370,8 +349,8 @@ class DataHandlerSingleStation(AbstractDataHandler):
:return: xarray.DataArray
"""
- self.data = self.data.interpolate_na(dim=dim, method=method, limit=limit, use_coordinate=use_coordinate,
- **kwargs)
+ self._data = self._data.interpolate_na(dim=dim, method=method, limit=limit, use_coordinate=use_coordinate,
+ **kwargs)
def make_history_window(self, dim_name_of_inputs: str, window: int, dim_name_of_shift: str) -> None:
"""
@@ -388,7 +367,8 @@ class DataHandlerSingleStation(AbstractDataHandler):
:param dim_name_of_shift: Dimension along shift will be applied
"""
window = -abs(window)
- self.history = self.shift(dim_name_of_shift, window).sel({dim_name_of_inputs: self.variables})
+ data = self.input_data.data
+ self.history = self.shift(data, dim_name_of_shift, window)
def make_labels(self, dim_name_of_target: str, target_var: str_or_list, dim_name_of_shift: str,
window: int) -> None:
@@ -404,7 +384,8 @@ class DataHandlerSingleStation(AbstractDataHandler):
:param window: lead time of label
"""
window = abs(window)
- self.label = self.shift(dim_name_of_shift, window).sel({dim_name_of_target: target_var})
+ data = self.target_data.data
+ self.label = self.shift(data, dim_name_of_shift, window)
def make_observation(self, dim_name_of_target: str, target_var: str_or_list, dim_name_of_shift: str) -> None:
"""
@@ -416,7 +397,8 @@ class DataHandlerSingleStation(AbstractDataHandler):
:param target_var: Name of observation variable(s) in 'dimension'
:param dim_name_of_shift: Name of dimension on which xarray.DataArray.shift will be applied
"""
- self.observation = self.shift(dim_name_of_shift, 0).sel({dim_name_of_target: target_var})
+ data = self.target_data.data
+ self.observation = self.shift(data, dim_name_of_shift, 0)
def remove_nan(self, dim: str) -> None:
"""
@@ -490,89 +472,23 @@ class DataHandlerSingleStation(AbstractDataHandler):
return data
@staticmethod
- def setup_transformation(transformation: Dict):
+ def setup_transformation(transformation: statistics.TransformationClass):
"""
Set up transformation by extracting all relevant information.
- Extract all information from transformation dictionary. Possible keys are method, mean, std, min, max.
- * If a transformation should be applied on base of existing values, these need to be provided in the respective
- keys "mean" and "std" (again only if required for given method).
-
- :param transformation: the transformation dictionary as described above.
-
- :return: updated transformation dictionary
-
- ## Transformation
-
- There are two different approaches (called scopes) to transform the data:
- 1) `station`: transform data for each station independently (somehow like batch normalisation)
- 1) `data`: transform all data of each station with shared metrics
-
- Transformation must be set by the `transformation` attribute. If `transformation = None` is given to `ExperimentSetup`,
- data is not transformed at all. For all other setups, use the following dictionary structure to specify the
- transformation.
- ```
- transformation = {"scope": <...>,
- "method": <...>,
- "mean": <...>,
- "std": <...>}
- ExperimentSetup(..., transformation=transformation, ...)
- ```
-
- ### scopes
-
- **station**: mean and std are not used
-
- **data**: either provide already calculated values for mean and std (if required by transformation method), or choose
- from different calculation schemes, explained in the mean and std section.
-
- ### supported transformation methods
- Currently supported methods are:
- * standardise (default, if method is not given)
- * centre
-
- ### mean and std
- `"mean"="accurate"`: calculate the accurate values of mean and std (depending on method) by using all data. Although,
- this method is accurate, it may take some time for the calculation. Furthermore, this could potentially lead to memory
- issue (not explored yet, but could appear for a very big amount of data)
-
- `"mean"="estimate"`: estimate mean and std (depending on method). For each station, mean and std are calculated and
- afterwards aggregated using the mean value over all station-wise metrics. This method is less accurate, especially
- regarding the std calculation but therefore much faster.
-
- We recommend to use the later method *estimate* because of following reasons:
- * much faster calculation
- * real accuracy of mean and std is less important, because it is "just" a transformation / scaling
- * accuracy of mean is almost as high as in the *accurate* case, because of
- $\bar{x_{ij}} = \bar{\left(\bar{x_i}\right)_j}$. The only difference is, that in the *estimate* case, each mean is
- equally weighted for each station independently of the actual data count of the station.
- * accuracy of std is lower for *estimate* because of $\var{x_{ij}} \ne \bar{\left(\var{x_i}\right)_j}$, but still the mean of all
- station-wise std is a decent estimate of the true std.
-
- `"mean"=<value, e.g. xr.DataArray>`: If mean and std are already calculated or shall be set manually, just add the
- scaling values instead of the calculation method. For method *centre*, std can still be None, but is required for the
- *standardise* method. **Important**: Format of given values **must** match internal data format of DataPreparation
- class: `xr.DataArray` with `dims=["variables"]` and one value for each variable.
-
+ * Either return new empty DataClass instances if given transformation arg is None,
+ * or return given object twice if transformation is a DataClass instance,
+ * or return the inputs and targets attributes if transformation is a TransformationClass instance (default
+ design behaviour)
"""
if transformation is None:
- return
- elif not isinstance(transformation, dict):
- raise TypeError(f"`transformation' must be either `None' or dict like e.g. `{{'method': 'standardise'}},"
- f" but transformation is of type {type(transformation)}.")
- transformation = transformation.copy()
- method = transformation.get("method", None)
- mean = transformation.get("mean", None)
- std = transformation.get("std", None)
- max_val = transformation.get("max", None)
- min_val = transformation.get("min", None)
-
- transformation["method"] = method
- transformation["mean"] = mean
- transformation["std"] = std
- transformation["max"] = max_val
- transformation["min"] = min_val
- return transformation
+ return statistics.DataClass(), statistics.DataClass()
+ elif isinstance(transformation, statistics.DataClass):
+ return transformation, transformation
+ elif isinstance(transformation, statistics.TransformationClass):
+ return copy.deepcopy(transformation.inputs), copy.deepcopy(transformation.targets)
+ else:
+ raise NotImplementedError("Cannot handle this.")
def load_data(self):
try:
@@ -581,8 +497,9 @@ class DataHandlerSingleStation(AbstractDataHandler):
self.download_data()
self.load_data()
- def transform(self, dim: Union[str, int] = 0, method: str = 'standardise', inverse: bool = False, mean=None,
- std=None, min_val=None, max_val=None) -> None:
+ def transform(self, data_class, dim: Union[str, int] = 0, transform_method: str = 'standardise',
+ inverse: bool = False, mean=None,
+ std=None, min=None, max=None) -> None:
"""
Transform data according to given transformation settings.
@@ -602,9 +519,9 @@ class DataHandlerSingleStation(AbstractDataHandler):
calculated over the data in this class instance.
:param std: Used for transformation (if required by 'method') based on external data. If 'None' the std is
calculated over the data in this class instance.
- :param min_val: Used for transformation (if required by 'method') based on external data. If 'None' min_val is
+ :param min: Used for transformation (if required by 'method') based on external data. If 'None' min_val is
extracted from the data in this class instance.
- :param max_val: Used for transformation (if required by 'method') based on external data. If 'None' max_val is
+ :param max: Used for transformation (if required by 'method') based on external data. If 'None' max_val is
extracted from the data in this class instance.
:return: xarray.DataArrays or pandas.DataFrames:
@@ -614,36 +531,37 @@ class DataHandlerSingleStation(AbstractDataHandler):
"""
def f(data):
- if method == 'standardise':
+ if transform_method == 'standardise':
return statistics.standardise(data, dim)
- elif method == 'centre':
+ elif transform_method == 'centre':
return statistics.centre(data, dim)
- elif method == 'normalise':
+ elif transform_method == 'normalise':
# use min/max of data or given min/max
raise NotImplementedError
else:
raise NotImplementedError
def f_apply(data):
- if method == "standardise":
+ if transform_method == "standardise":
return mean, std, statistics.standardise_apply(data, mean, std)
- elif method == "centre":
+ elif transform_method == "centre":
return mean, None, statistics.centre_apply(data, mean)
else:
raise NotImplementedError
if not inverse:
- if self._transform_method is not None:
- raise AssertionError(f"Transform method is already set. Therefore, data was already transformed with "
- f"{self._transform_method}. Please perform inverse transformation of data first.")
+ if data_class._method is not None:
+ raise AssertionError(f"Internal _method is already set. Therefore, data was already transformed with "
+ f"{data_class._method}. Please perform inverse transformation of data first.")
# apply transformation on local data instance (f) if mean is None, else apply by using mean (and std) from
# external data.
- self.mean, self.std, self.data = locals()["f" if mean is None else "f_apply"](self.data)
+ data_class.mean, data_class.std, data_class.data = locals()["f" if mean is None else "f_apply"](
+ data_class.data)
# set transform method to find correct method for inverse transformation.
- self._transform_method = method
+ data_class._method = transform_method
else:
- self.inverse_transform()
+ self.inverse_transform(data_class)
@staticmethod
def check_inverse_transform_params(mean: data_or_none, std: data_or_none, method: str) -> None:
@@ -665,7 +583,7 @@ class DataHandlerSingleStation(AbstractDataHandler):
if len(msg) > 0:
raise AttributeError(f"Inverse transform {method} can not be executed because following is None: {msg}")
- def inverse_transform(self) -> None:
+ def inverse_transform(self, data_class) -> None:
"""
Perform inverse transformation.
@@ -685,36 +603,26 @@ class DataHandlerSingleStation(AbstractDataHandler):
else:
raise NotImplementedError
- if self._transform_method is None:
+ if data_class.transform_method is None:
raise AssertionError("Inverse transformation method is not set. Data cannot be inverse transformed.")
- self.check_inverse_transform_params(self.mean, self.std, self._transform_method)
- self.data, self.mean, self.std = f_inverse(self.data, self.mean, self.std, self._transform_method)
- self._transform_method = None
+ self.check_inverse_transform_params(data_class.mean, data_class.std, data_class._method)
+ data_class.data, data_class.mean, data_class.std = f_inverse(data_class.data, data_class.mean, data_class.std,
+ data_class._method)
+ data_class.transform_method = None
# update X and Y
self.make_samples()
- def get_transformation_information(self, variable: str = None) -> Tuple[data_or_none, data_or_none, str]:
+ def get_transformation_targets(self) -> Tuple[data_or_none, data_or_none, str]:
"""
Extract transformation statistics and method.
- Get mean and standard deviation for given variable and the transformation method if set. If a transformation
+ Get mean and standard deviation for target values and the transformation method if set. If a transformation
depends only on particular statistics (e.g. only mean is required for centering), the remaining statistics are
returned with None as fill value.
- :param variable: Variable for which the information on transformation is requested.
-
:return: mean, standard deviation and transformation method
"""
- variable = self.target_var if variable is None else variable
- try:
- mean = self.mean.sel({'variables': variable}).values
- except AttributeError:
- mean = None
- try:
- std = self.std.sel({'variables': variable}).values
- except AttributeError:
- std = None
- return mean, std, self._transform_method
+ return self.target_data.mean, self.target_data.std, self.target_data.transform_method
if __name__ == "__main__":
@@ -727,7 +635,6 @@ if __name__ == "__main__":
time_dim='datetime', window_history_size=7, window_lead_time=3,
interpolation_limit=0
) # transformation={'method': 'standardise'})
- # sp.set_transformation({'method': 'standardise', 'mean': sp.mean+2, 'std': sp.std+1})
sp2 = DataHandlerSingleStation(data_path='/home/felix/PycharmProjects/mlt_new/data/', station='DEBY122',
statistics_per_var=statistics_per_var, station_type='background',
network='UBA', sampling='daily', target_dim='variables', target_var='o3',
diff --git a/mlair/data_handler/default_data_handler.py b/mlair/data_handler/default_data_handler.py
index 47f63a3e7bcbebd131c2a0da47d2e0833b02efed..e6dde10bf6bd13013fa454eadd1a7976c00dd3e2 100644
--- a/mlair/data_handler/default_data_handler.py
+++ b/mlair/data_handler/default_data_handler.py
@@ -4,6 +4,7 @@ __date__ = '2020-09-21'
import copy
import inspect
+import gc
import logging
import os
import pickle
@@ -15,7 +16,6 @@ import numpy as np
import xarray as xr
from mlair.data_handler.abstract_data_handler import AbstractDataHandler
-from mlair.data_handler.station_preparation import DataHandlerSingleStation
from mlair.helpers import remove_items, to_list
from mlair.helpers.join import EmptyQueryResult
@@ -25,11 +25,14 @@ num_or_list = Union[number, List[number]]
class DefaultDataHandler(AbstractDataHandler):
+ from mlair.data_handler.data_handler_single_station import DataHandlerSingleStation as data_handler
+ from mlair.data_handler.data_handler_single_station import DataHandlerSingleStation as data_handler_transformation
- _requirements = remove_items(inspect.getfullargspec(DataHandlerSingleStation).args, ["self", "station"])
+ _requirements = remove_items(inspect.getfullargspec(data_handler).args, ["self", "station"])
- def __init__(self, id_class: DataHandlerSingleStation, data_path: str, min_length: int = 0,
- extreme_values: num_or_list = None, extremes_on_right_tail_only: bool = False, name_affix=None):
+ def __init__(self, id_class: data_handler, data_path: str, min_length: int = 0,
+ extreme_values: num_or_list = None, extremes_on_right_tail_only: bool = False, name_affix=None,
+ store_processed_data=True):
super().__init__()
self.id_class = id_class
self.interpolation_dim = "datetime"
@@ -43,12 +46,12 @@ class DefaultDataHandler(AbstractDataHandler):
self._collection = self._create_collection()
self.harmonise_X()
self.multiply_extremes(extreme_values, extremes_on_right_tail_only, dim=self.interpolation_dim)
- self._store(fresh_store=True)
+ self._store(fresh_store=True, store_processed_data=store_processed_data)
@classmethod
def build(cls, station: str, **kwargs):
sp_keys = {k: copy.deepcopy(kwargs[k]) for k in cls._requirements if k in kwargs}
- sp = DataHandlerSingleStation(station, **sp_keys)
+ sp = cls.data_handler(station, **sp_keys)
dp_args = {k: copy.deepcopy(kwargs[k]) for k in cls.own_args("id_class") if k in kwargs}
return cls(sp, **dp_args)
@@ -61,6 +64,7 @@ class DefaultDataHandler(AbstractDataHandler):
def _reset_data(self):
self._X, self._Y, self._X_extreme, self._Y_extreme = None, None, None, None
+ gc.collect()
def _cleanup(self):
directory = os.path.dirname(self._save_file)
@@ -69,13 +73,14 @@ class DefaultDataHandler(AbstractDataHandler):
if os.path.exists(self._save_file):
shutil.rmtree(self._save_file, ignore_errors=True)
- def _store(self, fresh_store=False):
- self._cleanup() if fresh_store is True else None
- data = {"X": self._X, "Y": self._Y, "X_extreme": self._X_extreme, "Y_extreme": self._Y_extreme}
- with open(self._save_file, "wb") as f:
- pickle.dump(data, f)
- logging.debug(f"save pickle data to {self._save_file}")
- self._reset_data()
+ def _store(self, fresh_store=False, store_processed_data=True):
+ if store_processed_data is True:
+ self._cleanup() if fresh_store is True else None
+ data = {"X": self._X, "Y": self._Y, "X_extreme": self._X_extreme, "Y_extreme": self._Y_extreme}
+ with open(self._save_file, "wb") as f:
+ pickle.dump(data, f)
+ logging.debug(f"save pickle data to {self._save_file}")
+ self._reset_data()
def _load(self):
try:
@@ -140,7 +145,7 @@ class DefaultDataHandler(AbstractDataHandler):
return self.id_class.observation.copy().squeeze()
def get_transformation_Y(self):
- return self.id_class.get_transformation_information()
+ return self.id_class.get_transformation_targets()
def multiply_extremes(self, extreme_values: num_or_list = 1., extremes_on_right_tail_only: bool = False,
timedelta: Tuple[int, str] = (1, 'm'), dim="datetime"):
@@ -212,27 +217,55 @@ class DefaultDataHandler(AbstractDataHandler):
@classmethod
def transformation(cls, set_stations, **kwargs):
+ """
+ ### supported transformation methods
+
+ Currently supported methods are:
+
+ * standardise (default, if method is not given)
+ * centre
+
+ ### mean and std estimation
+
+ Mean and std (depending on method) are estimated. For each station, mean and std are calculated and afterwards
+ aggregated using the mean value over all station-wise metrics. This method is not exactly accurate, especially
+ regarding the std calculation but therefore much faster. Furthermore, it is a weighted mean weighted by the
+ time series length / number of data itself - a longer time series has more influence on the transformation
+ settings than a short time series. The estimation of the std in less accurate, because the unweighted mean of
+ all stds in not equal to the true std, but still the mean of all station-wise std is a decent estimate. Finally,
+ the real accuracy of mean and std is less important, because it is "just" a transformation / scaling.
+
+ ### mean and std given
+
+ 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.
+ """
+
sp_keys = {k: copy.deepcopy(kwargs[k]) for k in cls._requirements if k in kwargs}
- transformation_dict = sp_keys.pop("transformation")
- if transformation_dict is None:
+ transformation_class = sp_keys.get("transformation", None)
+ if transformation_class is None:
return
- scope = transformation_dict.pop("scope")
- method = transformation_dict.pop("method")
- if transformation_dict.pop("mean", None) is not None:
+
+ transformation_inputs = transformation_class.inputs
+ if transformation_inputs.mean is not None:
return
- mean, std = None, None
+ means = [None, None]
+ stds = [None, None]
for station in set_stations:
try:
- sp = DataHandlerSingleStation(station, transformation={"method": method}, **sp_keys)
- mean = sp.mean.copy(deep=True) if mean is None else mean.combine_first(sp.mean)
- std = sp.std.copy(deep=True) if std is None else std.combine_first(sp.std)
+ sp = cls.data_handler_transformation(station, **sp_keys)
+ for i, data in enumerate([sp.input_data, sp.target_data]):
+ means[i] = data.mean.copy(deep=True) if means[i] is None else means[i].combine_first(data.mean)
+ stds[i] = data.std.copy(deep=True) if stds[i] is None else stds[i].combine_first(data.std)
except (AttributeError, EmptyQueryResult):
continue
- if mean is None:
+ if means[0] is None:
return None
- mean_estimated = mean.mean("Stations")
- std_estimated = std.mean("Stations")
- return {"scope": scope, "method": method, "mean": mean_estimated, "std": std_estimated}
+ transformation_class.inputs.mean = means[0].mean("Stations")
+ transformation_class.inputs.std = stds[0].mean("Stations")
+ transformation_class.targets.mean = means[1].mean("Stations")
+ transformation_class.targets.std = stds[1].mean("Stations")
+ return transformation_class
def get_coordinates(self):
return self.id_class.get_coordinates()
\ No newline at end of file
diff --git a/mlair/helpers/join.py b/mlair/helpers/join.py
index a3c6876e3ea43ff4d03243430cf6cd791d62dec2..f66b277bbca54bd6190ab0430c9f8c0307b3f5af 100644
--- a/mlair/helpers/join.py
+++ b/mlair/helpers/join.py
@@ -55,7 +55,7 @@ def download_join(station_name: Union[str, List[str]], stat_var: dict, station_t
for var in _lower_list(sorted(vars_dict.keys())):
if var in stat_var.keys():
- logging.debug('load: {}'.format(var))
+ logging.debug('load: {}'.format(var)) # ToDo start here for #206
# create data link
opts = {'base': join_url_base, 'service': 'stats', 'id': vars_dict[var], 'statistics': stat_var[var],
@@ -138,6 +138,7 @@ def load_series_information(station_name: List[str], station_type: str_or_none,
opts = {"base": join_url_base, "service": "series", "station_id": station_name[0], "station_type": station_type,
"network_name": network_name}
station_vars = get_data(opts, headers)
+ logging.debug(f"{station_name}: {station_vars}") # ToDo start here for #206
vars_dict = {item[3].lower(): item[0] for item in station_vars}
return vars_dict
diff --git a/mlair/helpers/statistics.py b/mlair/helpers/statistics.py
index 056f92bec25b8d5216988f4dacb8fcd1e5257ab5..3db6618a5e8ebd575d61bc261144ff47ccaf9b53 100644
--- a/mlair/helpers/statistics.py
+++ b/mlair/helpers/statistics.py
@@ -9,10 +9,36 @@ import numpy as np
import xarray as xr
import pandas as pd
from typing import Union, Tuple, Dict
+from matplotlib import pyplot as plt
+
+from mlair.helpers import to_list, remove_items
Data = Union[xr.DataArray, pd.DataFrame]
+class DataClass:
+
+ def __init__(self, data=None, mean=None, std=None, max=None, min=None, transform_method=None):
+ self.data = data
+ self.mean = mean
+ self.std = std
+ self.max = max
+ self.min = min
+ self.transform_method = transform_method
+ self._method = None
+
+ def as_dict(self):
+ return remove_items(self.__dict__, "_method")
+
+
+class TransformationClass:
+
+ def __init__(self, inputs_mean=None, inputs_std=None, inputs_method=None, targets_mean=None, targets_std=None,
+ targets_method=None):
+ self.inputs = DataClass(mean=inputs_mean, std=inputs_std, transform_method=inputs_method)
+ self.targets = DataClass(mean=targets_mean, std=targets_std, transform_method=targets_method)
+
+
def apply_inverse_transformation(data: Data, mean: Data, std: Data = None, method: str = "standardise") -> Data:
"""
Apply inverse transformation for given statistics.
@@ -345,3 +371,168 @@ class SkillScores:
monthly_mean[monthly_mean.index.dt.month == month, :] = mu[mu.month == month].values
return monthly_mean
+
+
+class KolmogorovZurbenkoBaseClass:
+
+ def __init__(self, df, wl, itr, is_child=False, filter_dim="window"):
+ """
+ It create the variables associate with the Kolmogorov-Zurbenko-filter.
+
+ Args:
+ df(pd.DataFrame, None): time series of a variable
+ wl(list of int): window length
+ itr(list of int): number of iteration
+ """
+ self.df = df
+ self.filter_dim = filter_dim
+ self.wl = to_list(wl)
+ self.itr = to_list(itr)
+ if abs(len(self.wl) - len(self.itr)) > 0:
+ raise ValueError("Length of lists for wl and itr must agree!")
+ self._isChild = is_child
+ self.child = self.set_child()
+ self.type = type(self).__name__
+
+ def set_child(self):
+ if len(self.wl) > 1:
+ return KolmogorovZurbenkoBaseClass(None, self.wl[1:], self.itr[1:], True, self.filter_dim)
+ else:
+ return None
+
+ def kz_filter(self, df, m, k):
+ pass
+
+ def spectral_calc(self):
+ df_start = self.df
+ kz = self.kz_filter(df_start, self.wl[0], self.itr[0])
+ filtered = self.subtract(df_start, kz)
+ # case I: no child avail -> return kz and remaining
+ if self.child is None:
+ return [kz, filtered]
+ # case II: has child -> return current kz and all child results
+ else:
+ self.child.df = filtered
+ kz_next = self.child.spectral_calc()
+ return [kz] + kz_next
+
+ @staticmethod
+ def subtract(minuend, subtrahend):
+ try: # pandas implementation
+ return minuend.sub(subtrahend, axis=0)
+ except AttributeError: # general implementation
+ return minuend - subtrahend
+
+ def run(self):
+ return self.spectral_calc()
+
+ def transfer_function(self):
+ m = self.wl[0]
+ k = self.itr[0]
+ omega = np.linspace(0.00001, 0.15, 5000)
+ return omega, (np.sin(m * np.pi * omega) / (m * np.sin(np.pi * omega))) ** (2 * k)
+
+ def omega_null(self, alpha=0.5):
+ a = np.sqrt(6) / np.pi
+ b = 1 / (2 * np.array(self.itr))
+ c = 1 - alpha ** b
+ d = np.array(self.wl) ** 2 - alpha ** b
+ return a * np.sqrt(c / d)
+
+ def period_null(self, alpha=0.5):
+ return 1. / self.omega_null(alpha)
+
+ def period_null_days(self, alpha=0.5):
+ return self.period_null(alpha) / 24.
+
+ def plot_transfer_function(self, fig=None, name=None):
+ if fig is None:
+ fig = plt.figure()
+ omega, transfer_function = self.transfer_function()
+ if self.child is not None:
+ transfer_function_child = self.child.plot_transfer_function(fig)
+ else:
+ transfer_function_child = transfer_function * 0
+ plt.semilogx(omega, transfer_function - transfer_function_child,
+ label="m={:3.0f}, k={:3.0f}, T={:6.2f}d".format(self.wl[0],
+ self.itr[0],
+ self.period_null_days()))
+ plt.axvline(x=self.omega_null())
+ if not self._isChild:
+ locs, labels = plt.xticks()
+ plt.xticks(locs, np.round(1. / (locs * 24), 1))
+ plt.xlim([0.00001, 0.15])
+ plt.legend()
+ if name is None:
+ plt.show()
+ else:
+ plt.savefig(name)
+ else:
+ return transfer_function
+
+
+class KolmogorovZurbenkoFilterMovingWindow(KolmogorovZurbenkoBaseClass):
+
+ def __init__(self, df, wl: Union[list, int], itr: Union[list, int], is_child=False, filter_dim="window",
+ method="mean", percentile=0.5):
+ """
+ It create the variables associate with the KolmogorovZurbenkoFilterMovingWindow class.
+
+ Args:
+ df(pd.DataFrame, xr.DataArray): time series of a variable
+ wl: window length
+ itr: number of iteration
+ """
+ self.valid_methods = ["mean", "percentile", "median", "max", "min"]
+ if method not in self.valid_methods:
+ raise ValueError("Method '{}' is not supported. Please select from [{}].".format(
+ method, ", ".join(self.valid_methods)))
+ else:
+ self.method = method
+ if percentile > 1 or percentile < 0:
+ raise ValueError("Percentile must be in range [0, 1]. Given was {}!".format(percentile))
+ else:
+ self.percentile = percentile
+ super().__init__(df, wl, itr, is_child, filter_dim)
+
+ def set_child(self):
+ if len(self.wl) > 1:
+ return KolmogorovZurbenkoFilterMovingWindow(self.df, self.wl[1:], self.itr[1:], is_child=True,
+ filter_dim=self.filter_dim, method=self.method,
+ percentile=self.percentile)
+ else:
+ return None
+
+ def kz_filter(self, df, wl, itr):
+ """
+ It passes the low frequency time series.
+
+ Args:
+ wl(int): a window length
+ itr(int): a number of iteration
+ """
+ df_itr = df.__deepcopy__()
+ try:
+ kwargs = {"min_periods": 1,
+ "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()
+ for _ in np.arange(0, itr):
+ rolling = df_itr_var.rolling(**kwargs)
+ 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()
+ elif self.method == "min":
+ df_mv_avg_tmp = rolling.min()
+ 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)
+ return df_itr
+ except ValueError:
+ raise ValueError
diff --git a/mlair/model_modules/model_class.py b/mlair/model_modules/model_class.py
index c9cc13bd8108e43b5a9f03682942eacdf5a55f04..a603b466e4dab0dc30b6b6b22d10b6c27ee59767 100644
--- a/mlair/model_modules/model_class.py
+++ b/mlair/model_modules/model_class.py
@@ -396,8 +396,66 @@ class MyLittleModel(AbstractModelClass):
def set_compile_options(self):
self.initial_lr = 1e-2
self.optimizer = keras.optimizers.adam(lr=self.initial_lr)
- self.lr_decay = mlair.model_modules.keras_extensions.LearningRateDecay(base_lr=self.initial_lr, drop=.94,
- epochs_drop=10)
+ # self.lr_decay = mlair.model_modules.keras_extensions.LearningRateDecay(base_lr=self.initial_lr, drop=.94,
+ # epochs_drop=10)
+ self.compile_options = {"loss": [keras.losses.mean_squared_error], "metrics": ["mse", "mae"]}
+
+
+class MyLittleModelHourly(AbstractModelClass):
+ """
+ A customised model with a 1x1 Conv, and 4 Dense layers (64, 32, 16, window_lead_time), where the last layer is the
+ output layer depending on the window_lead_time parameter. Dropout is used between the Convolution and the first
+ Dense layer.
+ """
+
+ def __init__(self, input_shape: list, output_shape: list):
+ """
+ Sets model and loss depending on the given arguments.
+
+ :param shape_inputs: list of input shapes (expect len=1 with shape=(window_hist, station, variables))
+ :param shape_outputs: 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.dropout_rate = 0.1
+ self.regularizer = keras.regularizers.l2(0.001)
+ 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.
+ """
+
+ # add 1 to window_size to include current time step t0
+ x_input = keras.layers.Input(shape=self._input_shape)
+ x_in = keras.layers.Conv2D(128, (1, 1), padding='same', name='{}_Conv_1x1_128'.format("major"))(x_input)
+ x_in = self.activation()(x_in)
+ x_in = keras.layers.Conv2D(64, (1, 1), padding='same', name='{}_Conv_1x1_64'.format("major"))(x_in)
+ x_in = self.activation()(x_in)
+ x_in = keras.layers.Conv2D(32, (1, 1), padding='same', name='{}_Conv_1x1_32'.format("major"))(x_in)
+ x_in = self.activation()(x_in)
+ x_in = keras.layers.Flatten(name='{}'.format("major"))(x_in)
+ x_in = keras.layers.Dropout(self.dropout_rate, name='{}_Dropout_1'.format("major"))(x_in)
+ x_in = keras.layers.Dense(128, name='{}_Dense_128'.format("major"))(x_in)
+ x_in = self.activation()(x_in)
+ x_in = keras.layers.Dense(64, name='{}_Dense_64'.format("major"))(x_in)
+ x_in = self.activation()(x_in)
+ x_in = keras.layers.Dense(self._output_shape, name='{}_Dense'.format("major"))(x_in)
+ out_main = self.activation()(x_in)
+ self.model = keras.Model(inputs=x_input, outputs=[out_main])
+
+ def set_compile_options(self):
+ self.initial_lr = 1e-2
+ self.optimizer = keras.optimizers.SGD(lr=self.initial_lr, momentum=0.9)
self.compile_options = {"loss": [keras.losses.mean_squared_error], "metrics": ["mse", "mae"]}
diff --git a/mlair/plotting/postprocessing_plotting.py b/mlair/plotting/postprocessing_plotting.py
index 675e5ade587011a9ac835e9afb45f89173bc7653..c8682374e0d4c0d724d83a5e36977543ac3a50f8 100644
--- a/mlair/plotting/postprocessing_plotting.py
+++ b/mlair/plotting/postprocessing_plotting.py
@@ -137,15 +137,16 @@ class PlotMonthlySummary(AbstractPlotClass):
data_cnn = data.sel(type="CNN").squeeze()
if len(data_cnn.shape) > 1:
- data_cnn.coords["ahead"].values = [f"{days}d" for days in data_cnn.coords["ahead"].values]
+ data_cnn = data_cnn.assign_coords(ahead=[f"{days}d" for days in data_cnn.coords["ahead"].values])
data_obs = data.sel(type="obs", ahead=1).squeeze()
data_obs.coords["ahead"] = "obs"
data_concat = xr.concat([data_obs, data_cnn], dim="ahead")
- data_concat = data_concat.drop("type")
+ data_concat = data_concat.drop_vars("type")
- data_concat.index.values = data_concat.index.values.astype("datetime64[M]").astype(int) % 12 + 1
+ new_index = data_concat.index.values.astype("datetime64[M]").astype(int) % 12 + 1
+ data_concat = data_concat.assign_coords(index=new_index)
data_concat = data_concat.clip(min=0)
forecasts = xr.concat([forecasts, data_concat], 'index') if forecasts is not None else data_concat
@@ -902,6 +903,7 @@ class PlotAvailability(AbstractPlotClass):
# create standard Gantt plot for all stations (currently in single pdf file with single page)
super().__init__(plot_folder, "data_availability")
self.dim = time_dimension
+ self.linewidth = None
self.sampling = self._get_sampling(sampling)
plot_dict = self._prepare_data(generators)
lgd = self._plot(plot_dict)
@@ -917,11 +919,11 @@ class PlotAvailability(AbstractPlotClass):
lgd = self._plot(plot_dict_summary)
self._save(bbox_extra_artists=(lgd,), bbox_inches="tight")
- @staticmethod
- def _get_sampling(sampling):
+ def _get_sampling(self, sampling):
if sampling == "daily":
return "D"
elif sampling == "hourly":
+ self.linewidth = 0.001
return "h"
def _prepare_data(self, generators: Dict[str, DataCollection]):
@@ -982,7 +984,7 @@ class PlotAvailability(AbstractPlotClass):
plt_data = d.get(subset)
if plt_data is None:
continue
- ax.broken_barh(plt_data, (pos, height), color=color, edgecolor="white")
+ ax.broken_barh(plt_data, (pos, height), color=color, edgecolor="white", linewidth=self.linewidth)
yticklabels.append(station)
ax.set_ylim([height, number_of_stations + 1])
diff --git a/mlair/run_modules/post_processing.py b/mlair/run_modules/post_processing.py
index de43f30d929db1de12681d92c9c585df5c07944e..571d3a07d15873af1c1ccedc59e0cc462e07820f 100644
--- a/mlair/run_modules/post_processing.py
+++ b/mlair/run_modules/post_processing.py
@@ -399,10 +399,10 @@ class PostProcessing(RunEnvironment):
:return: filled data array with ols predictions
"""
tmp_ols = self.ols_model.predict(input_data)
- if not normalised:
- tmp_ols = statistics.apply_inverse_transformation(tmp_ols, mean, std, transformation_method)
target_shape = ols_prediction.values.shape
ols_prediction.values = np.swapaxes(tmp_ols, 2, 0) if target_shape != tmp_ols.shape else tmp_ols
+ if not normalised:
+ ols_prediction = statistics.apply_inverse_transformation(ols_prediction, mean, std, transformation_method)
return ols_prediction
def _create_persistence_forecast(self, data, persistence_prediction: xr.DataArray, mean: xr.DataArray,
@@ -423,9 +423,10 @@ class PostProcessing(RunEnvironment):
:return: filled data array with persistence predictions
"""
tmp_persi = data.copy()
- if not normalised:
- tmp_persi = statistics.apply_inverse_transformation(tmp_persi, mean, std, transformation_method)
persistence_prediction.values = np.tile(tmp_persi, (self.window_lead_time, 1)).T
+ if not normalised:
+ persistence_prediction = statistics.apply_inverse_transformation(persistence_prediction, mean, std,
+ transformation_method)
return persistence_prediction
def _create_nn_forecast(self, input_data: xr.DataArray, nn_prediction: xr.DataArray, mean: xr.DataArray,
@@ -447,8 +448,6 @@ class PostProcessing(RunEnvironment):
:return: filled data array with nn predictions
"""
tmp_nn = self.model.predict(input_data)
- if not normalised:
- tmp_nn = statistics.apply_inverse_transformation(tmp_nn, mean, std, transformation_method)
if isinstance(tmp_nn, list):
nn_prediction.values = tmp_nn[-1]
elif tmp_nn.ndim == 3:
@@ -457,6 +456,8 @@ class PostProcessing(RunEnvironment):
nn_prediction.values = tmp_nn
else:
raise NotImplementedError(f"Number of dimension of model output must be 2 or 3, but not {tmp_nn.dims}.")
+ if not normalised:
+ nn_prediction = statistics.apply_inverse_transformation(nn_prediction, mean, std, transformation_method)
return nn_prediction
@staticmethod
@@ -528,7 +529,7 @@ class PostProcessing(RunEnvironment):
# external_data = external_data.squeeze("Stations").sel(window=1).drop(["window", "Stations", "variables"])
external_data = self._create_observation(observation, None, mean, std, transformation_method, normalised=False)
return external_data.rename({external_data.dims[0]: 'index'})
- except IndexError:
+ except (IndexError, KeyError):
return None
def calculate_skill_scores(self) -> Tuple[Dict, Dict]:
diff --git a/mlair/run_modules/pre_processing.py b/mlair/run_modules/pre_processing.py
index ed972896e7a39b0b56df23dbc8a8d1ae64fb4183..82af9cf02cda9401237bac15ccf0a52fa10acdad 100644
--- a/mlair/run_modules/pre_processing.py
+++ b/mlair/run_modules/pre_processing.py
@@ -56,7 +56,8 @@ class PreProcessing(RunEnvironment):
def _run(self):
stations = self.data_store.get("stations")
data_handler = self.data_store.get("data_handler")
- _, valid_stations = self.validate_station(data_handler, stations, "preprocessing", overwrite_local_data=True)
+ _, valid_stations = self.validate_station(data_handler, stations,
+ "preprocessing") # , store_processed_data=False)
if len(valid_stations) == 0:
raise ValueError("Couldn't find any valid data according to given parameters. Abort experiment run.")
self.data_store.set("stations", valid_stations)
@@ -192,20 +193,14 @@ class PreProcessing(RunEnvironment):
self.data_store.set("stations", valid_stations, scope=set_name)
self.data_store.set("data_collection", collection, scope=set_name)
- def validate_station(self, data_handler: AbstractDataHandler, set_stations, set_name=None, overwrite_local_data=False):
+ def validate_station(self, data_handler: AbstractDataHandler, set_stations, set_name=None,
+ store_processed_data=True):
"""
Check if all given stations in `all_stations` are valid.
Valid means, that there is data available for the given time range (is included in `kwargs`). The shape and the
loading time are logged in debug mode.
- :param args: Dictionary with required parameters for DataGenerator class (`data_path`, `network`, `stations`,
- `variables`, `time_dim`, `target_dim`, `target_var`).
- :param kwargs: positional parameters for the DataGenerator class (e.g. `start`, `interpolation_method`,
- `window_lead_time`).
- :param all_stations: All stations to check.
- :param name: name to display in the logging info message
-
:return: Corrected list containing only valid station IDs.
"""
t_outer = TimeTracking()
@@ -219,7 +214,8 @@ class PreProcessing(RunEnvironment):
kwargs = self.data_store.create_args_dict(data_handler.requirements(), scope=set_name)
for station in set_stations:
try:
- dp = data_handler.build(station, name_affix=set_name, **kwargs)
+ dp = data_handler.build(station, name_affix=set_name, store_processed_data=store_processed_data,
+ **kwargs)
collection.add(dp)
valid_stations.append(station)
except (AttributeError, EmptyQueryResult):
@@ -234,6 +230,3 @@ class PreProcessing(RunEnvironment):
transformation_dict = data_handler.transformation(stations, **kwargs)
if transformation_dict is not None:
self.data_store.set("transformation", transformation_dict)
-
-
-
diff --git a/mlair/workflows/abstract_workflow.py b/mlair/workflows/abstract_workflow.py
index bced90bbe848cc9ebe36c583d05b62549f0ae80b..3a627d9f72a5c1c97c35b464af1b0944bc397ea5 100644
--- a/mlair/workflows/abstract_workflow.py
+++ b/mlair/workflows/abstract_workflow.py
@@ -16,15 +16,17 @@ class Workflow:
execution but not the dependencies (workflow would probably fail in this case)."""
def __init__(self, name=None):
- self._registry = OrderedDict()
+ self._registry_kwargs = {}
+ self._registry = []
self._name = name if name is not None else self.__class__.__name__
def add(self, stage, **kwargs):
"""Add a new stage with optional kwargs."""
- self._registry[stage] = kwargs
+ self._registry.append(stage)
+ self._registry_kwargs[len(self._registry) - 1] = kwargs
def run(self):
"""Run workflow embedded in a run environment and according to the stage's ordering."""
with RunEnvironment(name=self._name):
- for stage, kwargs in self._registry.items():
- stage(**kwargs)
+ for pos, stage in enumerate(self._registry):
+ stage(**self._registry_kwargs[pos])
diff --git a/mlair/workflows/default_workflow.py b/mlair/workflows/default_workflow.py
index 85d6726b70b699968933bf9af7580895490b8a6d..4d113190fdc90ec852d7db2b33459b9162867a24 100644
--- a/mlair/workflows/default_workflow.py
+++ b/mlair/workflows/default_workflow.py
@@ -14,28 +14,29 @@ class DefaultWorkflow(Workflow):
the mentioned ordering."""
def __init__(self, stations=None,
- train_model=None, create_new_model=None,
- window_history_size=None,
- experiment_date="testrun",
- variables=None, statistics_per_var=None,
- start=None, end=None,
- target_var=None, target_dim=None,
- window_lead_time=None,
- dimensions=None,
- interpolation_method=None, time_dim=None, limit_nan_fill=None,
- train_start=None, train_end=None, val_start=None, val_end=None, test_start=None, test_end=None,
- use_all_stations_on_all_data_sets=None, fraction_of_train=None,
- experiment_path=None, plot_path=None, forecast_path=None, bootstrap_path=None, overwrite_local_data=None,
- sampling=None,
- permute_data_on_training=None, extreme_values=None, extremes_on_right_tail_only=None,
- transformation=None,
- train_min_length=None, val_min_length=None, test_min_length=None,
- evaluate_bootstraps=None, number_of_bootstraps=None, create_new_bootstraps=None,
- plot_list=None,
- model=None,
- batch_size=None,
- epochs=None,
- data_preparation=None,
+ train_model=None, create_new_model=None,
+ window_history_size=None,
+ experiment_date="testrun",
+ variables=None, statistics_per_var=None,
+ start=None, end=None,
+ target_var=None, target_dim=None,
+ window_lead_time=None,
+ dimensions=None,
+ interpolation_method=None, time_dim=None, limit_nan_fill=None,
+ train_start=None, train_end=None, val_start=None, val_end=None, test_start=None, test_end=None,
+ use_all_stations_on_all_data_sets=None, fraction_of_train=None,
+ experiment_path=None, plot_path=None, forecast_path=None, bootstrap_path=None,
+ overwrite_local_data=None,
+ sampling=None,
+ permute_data_on_training=None, extreme_values=None, extremes_on_right_tail_only=None,
+ transformation=None,
+ train_min_length=None, val_min_length=None, test_min_length=None,
+ evaluate_bootstraps=None, number_of_bootstraps=None, create_new_bootstraps=None,
+ plot_list=None,
+ model=None,
+ batch_size=None,
+ epochs=None,
+ data_handler=None,
**kwargs):
super().__init__()
@@ -58,28 +59,29 @@ class DefaultWorkflowHPC(Workflow):
Training and PostProcessing in exact the mentioned ordering."""
def __init__(self, stations=None,
- train_model=None, create_new_model=None,
- window_history_size=None,
- experiment_date="testrun",
- variables=None, statistics_per_var=None,
- start=None, end=None,
- target_var=None, target_dim=None,
- window_lead_time=None,
- dimensions=None,
- interpolation_method=None, time_dim=None, limit_nan_fill=None,
- train_start=None, train_end=None, val_start=None, val_end=None, test_start=None, test_end=None,
- use_all_stations_on_all_data_sets=None, fraction_of_train=None,
- experiment_path=None, plot_path=None, forecast_path=None, bootstrap_path=None, overwrite_local_data=None,
- sampling=None,
- permute_data_on_training=None, extreme_values=None, extremes_on_right_tail_only=None,
- transformation=None,
- train_min_length=None, val_min_length=None, test_min_length=None,
- evaluate_bootstraps=None, number_of_bootstraps=None, create_new_bootstraps=None,
- plot_list=None,
- model=None,
- batch_size=None,
- epochs=None,
- data_preparation=None, **kwargs):
+ train_model=None, create_new_model=None,
+ window_history_size=None,
+ experiment_date="testrun",
+ variables=None, statistics_per_var=None,
+ start=None, end=None,
+ target_var=None, target_dim=None,
+ window_lead_time=None,
+ dimensions=None,
+ interpolation_method=None, time_dim=None, limit_nan_fill=None,
+ train_start=None, train_end=None, val_start=None, val_end=None, test_start=None, test_end=None,
+ use_all_stations_on_all_data_sets=None, fraction_of_train=None,
+ experiment_path=None, plot_path=None, forecast_path=None, bootstrap_path=None,
+ overwrite_local_data=None,
+ sampling=None,
+ permute_data_on_training=None, extreme_values=None, extremes_on_right_tail_only=None,
+ transformation=None,
+ train_min_length=None, val_min_length=None, test_min_length=None,
+ evaluate_bootstraps=None, number_of_bootstraps=None, create_new_bootstraps=None,
+ plot_list=None,
+ model=None,
+ batch_size=None,
+ epochs=None,
+ data_handler=None, **kwargs):
super().__init__()
# extract all given kwargs arguments
diff --git a/requirements.txt b/requirements.txt
index be76eab5b74797b039682a292ae8890488c058ec..371bb776e581925e507bf06c60bd866061c52791 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -61,7 +61,7 @@ typing-extensions
urllib3==1.25.8
wcwidth==0.1.8
Werkzeug==1.0.0
-xarray==0.15.0
+xarray==0.16.1
zipp==3.1.0
setuptools~=49.6.0
diff --git a/test/test_configuration/test_defaults.py b/test/test_configuration/test_defaults.py
new file mode 100644
index 0000000000000000000000000000000000000000..7dc7199f2d8ed75af2d4f968a1f52ff3ee15baec
--- /dev/null
+++ b/test/test_configuration/test_defaults.py
@@ -0,0 +1,73 @@
+from mlair.configuration.defaults import *
+
+
+class TestGetDefaults:
+
+ def test_get_defaults(self):
+ defaults = get_defaults()
+ assert isinstance(defaults, dict)
+ assert all(map(lambda k: k in defaults.keys(), ["DEFAULT_STATIONS", "DEFAULT_BATCH_SIZE", "DEFAULT_PLOT_LIST"]))
+ assert all(map(lambda x: x.startswith("DEFAULT"), defaults.keys()))
+
+
+class TestAllDefaults:
+
+ def test_training_parameters(self):
+ assert DEFAULT_CREATE_NEW_MODEL is True
+ assert DEFAULT_TRAIN_MODEL is True
+ assert DEFAULT_FRACTION_OF_TRAINING == 0.8
+ assert DEFAULT_EXTREME_VALUES is None
+ assert DEFAULT_EXTREMES_ON_RIGHT_TAIL_ONLY is False
+ assert DEFAULT_PERMUTE_DATA is False
+ assert DEFAULT_BATCH_SIZE == int(256 * 2)
+ assert DEFAULT_EPOCHS == 20
+
+ def test_data_handler_parameters(self):
+ assert DEFAULT_STATIONS == ['DEBW107', 'DEBY081', 'DEBW013', 'DEBW076', 'DEBW087']
+ assert DEFAULT_VAR_ALL_DICT == {'o3': 'dma8eu', 'relhum': 'average_values', 'temp': 'maximum',
+ 'u': 'average_values',
+ 'v': 'average_values', 'no': 'dma8eu', 'no2': 'dma8eu',
+ 'cloudcover': 'average_values',
+ 'pblheight': 'maximum'}
+ assert DEFAULT_NETWORK == "AIRBASE"
+ assert DEFAULT_STATION_TYPE == "background"
+ assert DEFAULT_VARIABLES == DEFAULT_VAR_ALL_DICT.keys()
+ assert DEFAULT_START == "1997-01-01"
+ assert DEFAULT_END == "2017-12-31"
+ assert DEFAULT_WINDOW_HISTORY_SIZE == 13
+ assert DEFAULT_OVERWRITE_LOCAL_DATA is False
+ assert isinstance(DEFAULT_TRANSFORMATION, TransformationClass)
+ assert DEFAULT_TRANSFORMATION.inputs.transform_method == "standardise"
+ assert DEFAULT_TRANSFORMATION.targets.transform_method == "standardise"
+ assert DEFAULT_TARGET_VAR == "o3"
+ assert DEFAULT_TARGET_DIM == "variables"
+ assert DEFAULT_WINDOW_LEAD_TIME == 3
+ assert DEFAULT_DIMENSIONS == {"new_index": ["datetime", "Stations"]}
+ assert DEFAULT_TIME_DIM == "datetime"
+ assert DEFAULT_INTERPOLATION_METHOD == "linear"
+ assert DEFAULT_INTERPOLATION_LIMIT == 1
+
+ def test_subset_parameters(self):
+ assert DEFAULT_TRAIN_START == "1997-01-01"
+ assert DEFAULT_TRAIN_END == "2007-12-31"
+ assert DEFAULT_TRAIN_MIN_LENGTH == 90
+ assert DEFAULT_VAL_START == "2008-01-01"
+ assert DEFAULT_VAL_END == "2009-12-31"
+ assert DEFAULT_VAL_MIN_LENGTH == 90
+ assert DEFAULT_TEST_START == "2010-01-01"
+ assert DEFAULT_TEST_END == "2017-12-31"
+ assert DEFAULT_TEST_MIN_LENGTH == 90
+ assert DEFAULT_TRAIN_VAL_MIN_LENGTH == 180
+ assert DEFAULT_USE_ALL_STATIONS_ON_ALL_DATA_SETS is True
+
+ def test_hpc_parameters(self):
+ assert DEFAULT_HPC_HOST_LIST == ["jw", "hdfmlc"]
+ assert DEFAULT_HPC_LOGIN_LIST == ["ju", "hdfmll"]
+
+ def test_postprocessing_parameters(self):
+ assert DEFAULT_EVALUATE_BOOTSTRAPS is True
+ assert DEFAULT_CREATE_NEW_BOOTSTRAPS is False
+ assert DEFAULT_NUMBER_OF_BOOTSTRAPS == 20
+ assert DEFAULT_PLOT_LIST == ["PlotMonthlySummary", "PlotStationMap", "PlotClimatologicalSkillScore",
+ "PlotTimeSeries", "PlotCompetitiveSkillScore", "PlotBootstrapSkillScore",
+ "PlotConditionalQuantiles", "PlotAvailability"]
diff --git a/test/test_statistics.py b/test/test_statistics.py
index d4a72674ae89ecd106ff1861aa6ee26567da3243..76adc1bdd210e072b4fc9be717269c6ceb951fec 100644
--- a/test/test_statistics.py
+++ b/test/test_statistics.py
@@ -3,7 +3,9 @@ import pandas as pd
import pytest
import xarray as xr
-from mlair.helpers.statistics import standardise, standardise_inverse, standardise_apply, centre, centre_inverse, centre_apply, \
+from mlair.helpers.statistics import DataClass, TransformationClass
+from mlair.helpers.statistics import standardise, standardise_inverse, standardise_apply, centre, centre_inverse, \
+ centre_apply, \
apply_inverse_transformation
lazy = pytest.lazy_fixture
@@ -113,3 +115,50 @@ class TestCentre:
data = centre_apply(data_orig, mean)
mean_expected = np.array([2, -5, 10]) - np.array([2, 10, 3])
assert np.testing.assert_almost_equal(data.mean(dim), mean_expected, decimal=1) is None
+
+
+class TestDataClass:
+
+ def test_init(self):
+ dc = DataClass()
+ assert all([obj is None for obj in [dc.data, dc.mean, dc.std, dc.max, dc.min, dc.transform_method, dc._method]])
+
+ def test_init_values(self):
+ dc = DataClass(data=12, mean=2, std="test", max=23.4, min=np.array([3]), transform_method="f")
+ assert dc.data == 12
+ assert dc.mean == 2
+ assert dc.std == "test"
+ assert dc.max == 23.4
+ assert np.testing.assert_array_equal(dc.min, np.array([3])) is None
+ assert dc.transform_method == "f"
+ assert dc._method is None
+
+ def test_as_dict(self):
+ dc = DataClass(std=23)
+ dc._method = "f(x)"
+ assert dc.as_dict() == {"data": None, "mean": None, "std": 23, "max": None, "min": None,
+ "transform_method": None}
+
+
+class TestTransformationClass:
+
+ def test_init(self):
+ tc = TransformationClass()
+ assert hasattr(tc, "inputs")
+ assert isinstance(tc.inputs, DataClass)
+ assert hasattr(tc, "targets")
+ assert isinstance(tc.targets, DataClass)
+ assert tc.inputs.mean is None
+ assert tc.targets.std is None
+
+ def test_init_values(self):
+ tc = TransformationClass(inputs_mean=1, inputs_std=2, inputs_method="f", targets_mean=3, targets_std=4,
+ targets_method="g")
+ assert tc.inputs.mean == 1
+ assert tc.inputs.std == 2
+ assert tc.inputs.transform_method == "f"
+ assert tc.inputs.max is None
+ assert tc.targets.mean == 3
+ assert tc.targets.std == 4
+ assert tc.targets.transform_method == "g"
+ assert tc.inputs.min is None