Coverage for src/lsqfitgp/_GP/__init__.py: 100%

1# lsqfitgp/_GP/__init__.py 

2# 

3# Copyright (c) 2023, Giacomo Petrillo 

4# 

5# This file is part of lsqfitgp. 

6# 

7# lsqfitgp is free software: you can redistribute it and/or modify 

8# it under the terms of the GNU General Public License as published by 

9# the Free Software Foundation, either version 3 of the License, or 

10# (at your option) any later version. 

11# 

12# lsqfitgp is distributed in the hope that it will be useful, 

13# but WITHOUT ANY WARRANTY; without even the implied warranty of 

14# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 

15# GNU General Public License for more details. 

16# 

17# You should have received a copy of the GNU General Public License 

18# along with lsqfitgp. If not, see <http://www.gnu.org/licenses/>. 

19 

20from ._gp import GP 6 ctx1abcdef

21 

22# TODO planned rewrite: 

23# Methods: 

24# Processes 

25# .defproc(key, *args, *, deriv=0) 

26# *args is either (mean, kernel) or just (kernel,). The mean is an 

27# ordinary callable. 

28# Inspection 

29# .elements() return a read-only mapping of keys -> named tuples with 

30# fields ancestors=(tuple of keys), kind=str (x, cov, transf), shape, 

31# x (array or None), proc (proc key or None). The mapping generates 

32# the descriptions on the fly in __getitem__. 

33# .processes() analogous with ancestors, kind (kernel, transf), kernel 

34# (Kernel or None) 

35# .dtype the x dtype, None if no x passed yet 

36# .kernel(*args) if one key a Kernel, if two a CrossKernel 

37# .meanfunc(prockey) 

38# .__repr__() -> multiline readable description 

39# Mean functions 

40# .__init__(*args, ...) supports either only the kernel or the mean and 

41# the kernel of the DefaultProcess. 

42# .add(*args) arrays/dicts mean and cov or just cov, replaces .addcov. 

43# Conditioning 

44# .condition(given, givencov=None) -> new GP, with appended new keys and 

45# values to condition on. given and givencov keep dict layout like 

46# now. 

47# .extend(given, givencov=None) -> new GP, like current predfromfit. It 

48# sets a flag about the kind of conditioning in the list of given 

49# stuff. I have to derive how to condition and extend together. 

50# Distribution access 

51# .mean(key) -> array, the key can be a pytree 

52# .cov(*args) -> if one arg: square covariance, if two args: cross 

53# covariance. The keys can be pytrees, and the result is the pytree 

54# product with matrices as leaves. 

55# .gvars(key, keepcorr=True) -> tree of arrays of gvars. keepcorr=True 

56# won't work if condition was not passed gvars, so this default raises 

57# an error if that happens. 

58# .sample(key, n=1) -> array, key can be a pytree. caches the decomp? 

59# Density 

60# ._decomp(y, ycov=None) -> resid, cov (like _prior_decomp now) 

61# .logpdf(y, ycov=None) -> like marginal_likelihood now 

62 

63# TODO methods to implement linear models. The elegant way to do this is with 

64# kernels taking a formula, but that would not take advantage of the fact that 

65# the kernel is low-rank if I re-implement Woodbury. I need to define an element 

66# for the coefficients, and then the process works by definining finite 

67# transformations of this element instead of going through the usual kernel 

68# route. 

69# 

70# A more general route would be `features` mechanism for kernels: a kernel can 

71# provide itself a way to evaluate a covariance matrix as a low-rank product.