-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathchangelog.txt
executable file
·72 lines (48 loc) · 2.95 KB
/
changelog.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
# Changelog
0.2.1 (working release)
-----------------------
1. Try JET or forward Laplacian for efficient computation of recursive derivatives.
0.2.0 (31 October 2024)
-----------------------
1. Deprecated the old `setup.py`; now use `pyproject.toml`.
2. Added semantic typings for better documentation.
3. Added support for SDEs with time-dependent drift/dispersion, as well as for target function \phi. The time variable support is not implemented for `tme.mean_and_cov` as we don't have a consistent approximation for the covariance part for now. Note implemented for the `sympy` module either.
4. Migrated all the `unittest` to `pytest`.
0.1.5 (8 October, 2022)
-----------------------
1. Fixed a critical bug in computing the matrix-Hessian-matrix multiplication. Fortunately this bug does not affect the results when using constant dispersion coefficient.
0.1.4 (8 June, 2022)
-----------------------
1. Changed the verbose printing of TME matlab.
2. Updated docs by adding TME filters and smoothers.
3. Changed the function signatures `a` and `b` to `drift` and `dispersion` to be more explicit.
4. Removed the spectral density `Qw`. This `Qw` seems to be useful in some isolated cases only. If one wants to specify `Qw`, please put its Cholesky decomposition into the dispersion function. Note that this change does not affect the Matlab implementation.
5. Some code simplifications.
6. Fixed a plotting error in `examples/generate_lorenz_anime.py`. This error is due to a bug of `matplotlib`, see https://github.com/matplotlib/matplotlib/issues/22308.
7. The `sympy` simplification problem seems to be solved in the current version 1.10.1.
...
0.1.3 (17 August, 2021)
-----------------------
1. Added a new Lorenz example as well as some documentations.
2. The requirement for Python is now down to 3.7 (JaX does not support 3.6 anymore). The necessity for using `math.comb` is removed.
0.1.2 (16 August, 2021)
-----------------------
1. Removed `_phi_i`, `_phi_ii`, and the naive implementation `generator_power_naive` in `base_jax`.
The function `generator` is also written as a special case of `generator_power` when `order=1`. The old
implementaions are now moved to the test folder.
2. Make sure the shape of `b` and `Qw` are consistent. Added `_format_noise` and `_format_dispersion`.
3. Removed the requirements for jax in `requirement.txt`. The users must install jax by themself so that
they can choose to use cpu or gpu.
4. Slightly updated the examples in `./python/examples`.
5. In the docstrings of `base_jax` there were wrong "Symbolic spectral ...". Removed.
6. Documentation updated.
0.1.1 (15 August, 2021)
-----------------------
1. Examples in README.md and ./python/README.md are missing `import jax.numpy as jnp`.
Fixed.
2. Rewrote the Hessian product in `base_jax.generator_power()`. The unnecessary argument `phi_out_ndims`
is now removed.
3. Added unittest for 1d and 2d target function phi
0.1.0 (15 August, 2021)
-----------------------
Initial release.