|
| 1 | +""" |
| 2 | +This file is part of CLIMADA. |
| 3 | +
|
| 4 | +Copyright (C) 2017 ETH Zurich, CLIMADA contributors listed in AUTHORS. |
| 5 | +
|
| 6 | +CLIMADA is free software: you can redistribute it and/or modify it under the |
| 7 | +terms of the GNU General Public License as published by the Free |
| 8 | +Software Foundation, version 3. |
| 9 | +
|
| 10 | +CLIMADA is distributed in the hope that it will be useful, but WITHOUT ANY |
| 11 | +WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A |
| 12 | +PARTICULAR PURPOSE. See the GNU General Public License for more details. |
| 13 | +
|
| 14 | +You should have received a copy of the GNU General Public License along |
| 15 | +with CLIMADA. If not, see <https://www.gnu.org/licenses/>. |
| 16 | +
|
| 17 | +--- |
| 18 | +
|
| 19 | +Define Forecast base class. |
| 20 | +""" |
| 21 | + |
| 22 | +from abc import ABC, abstractmethod |
| 23 | +from typing import Literal, Self |
| 24 | + |
| 25 | +import numpy as np |
| 26 | +from scipy.sparse import block_array, csr_matrix |
| 27 | + |
| 28 | + |
| 29 | +def reduce_mat(mat: csr_matrix, func): |
| 30 | + """Reduce a matrix and return the CSR representation""" |
| 31 | + return func(mat).tocsr() |
| 32 | + |
| 33 | + |
| 34 | +def concat_matrices_per_event(*matrices: csr_matrix): |
| 35 | + """Concatenate matrices by event""" |
| 36 | + return block_array([[mat] for mat in matrices], format="csr") |
| 37 | + |
| 38 | + |
| 39 | +def matrix_event_padding(mat: csr_matrix, num_events): |
| 40 | + """Pad zero events""" |
| 41 | + pad_events = mat.shape[0] - num_events |
| 42 | + if pad_events < 1: |
| 43 | + return mat |
| 44 | + return block_array( |
| 45 | + [[mat], csr_matrix((pad_events, mat.shape[1], mat.dtype))], format="csr" |
| 46 | + ) |
| 47 | + |
| 48 | + |
| 49 | +class Forecast(ABC): |
| 50 | + lead_time: np.ndarray[np.timedelta64] |
| 51 | + member: np.ndarray[int] |
| 52 | + forecast_date: np.datetime64 | None |
| 53 | + num_members: int |
| 54 | + num_lead_times: int |
| 55 | + |
| 56 | + def __init__(self, lead_time, member, forecast_date: np.datetime64 | None = None): |
| 57 | + """Store members""" |
| 58 | + pass |
| 59 | + |
| 60 | + # --- Selection --- # |
| 61 | + |
| 62 | + @abstractmethod |
| 63 | + def _select_by_index(self, index: tuple[np.ndarray, ...]) -> Self: |
| 64 | + """Return a new object with the index used for selecting events""" |
| 65 | + raise NotImplementedError |
| 66 | + |
| 67 | + def _select_member(self, member: int | None) -> np.ndarray: |
| 68 | + """Return boolean array where self.member == member""" |
| 69 | + ... |
| 70 | + |
| 71 | + def _select_lead_time(self, lead_time: np.timedelta64 | None) -> np.ndarray: |
| 72 | + """Return boolean array where self.lead_time == lead_time""" |
| 73 | + ... |
| 74 | + |
| 75 | + def select(self, *, member: int | None, lead_time: np.timedelta64 | None) -> Self: |
| 76 | + index = np.nonzero( |
| 77 | + self._select_member(member) & self._select_lead_time(lead_time) |
| 78 | + ) |
| 79 | + return self._select_by_index(index) |
| 80 | + |
| 81 | + # --- Generic reduction --- # |
| 82 | + |
| 83 | + @classmethod |
| 84 | + @abstractmethod |
| 85 | + def concat(cls, *obj: Self) -> Self: |
| 86 | + """Concatenate multiple object instances""" |
| 87 | + raise NotImplementedError |
| 88 | + |
| 89 | + @abstractmethod |
| 90 | + def _reduce(self, func) -> Self: |
| 91 | + """Apply the reduction function in the derived class and return the result |
| 92 | +
|
| 93 | + Note: The derived class will likely need to pad matrices! |
| 94 | + """ |
| 95 | + raise NotImplementedError |
| 96 | + |
| 97 | + def reduce(self, func, dim: Literal["member", "lead_time"] | None = None) -> Self: |
| 98 | + """Reduce along a given dimension with func""" |
| 99 | + if dim is None: |
| 100 | + # TODO: Check if we selected a specific member or lead time. |
| 101 | + # Pad events accordingly! |
| 102 | + return self._reduce(func=func) # Derived class specialization |
| 103 | + |
| 104 | + return self.concat( |
| 105 | + *( |
| 106 | + self.select(**{dim: val}).reduce(func=func, dim=None) |
| 107 | + for val in np.unique(getattr(self, dim)) |
| 108 | + ) |
| 109 | + ) |
| 110 | + |
| 111 | + # --- Specializations --- # |
| 112 | + |
| 113 | + @abstractmethod |
| 114 | + def _max(self) -> Self: |
| 115 | + """Apply the maximum function in the derived class and return the result""" |
| 116 | + raise NotImplementedError |
| 117 | + |
| 118 | + def _reduce_attr( |
| 119 | + self, attr: str, dim: Literal["member", "lead_time"] | None = None |
| 120 | + ) -> Self: |
| 121 | + """Reduce along a given dimension with attribute attr""" |
| 122 | + if dim is None: |
| 123 | + # TODO: Check if we selected a specific member or lead time. |
| 124 | + # Pad events accordingly! |
| 125 | + return getattr(self, "_" + attr)() # Derived class specialization |
| 126 | + |
| 127 | + return self.concat( |
| 128 | + *( |
| 129 | + getattr(self.select(**{dim: val}), attr)(dim=None) |
| 130 | + for val in np.unique(getattr(self, dim)) |
| 131 | + ) |
| 132 | + ) |
| 133 | + |
| 134 | + def max(self, dim): |
| 135 | + return self._reduce_attr(attr="max", dim=dim) |
0 commit comments