Improve accuracy for np.log(1 + ... )

[hai-dang-dau]

In the file "./particles/resampling.py", line 272: the use of np.log(1 + ...) should be replaced by np.log1p(...). The documentation of numpy says that np.log1p(x) is more accurate than np.log(1+x). Here is my experiment:

np.log(1 + np.exp(-3))
Out[66]: 0.04858735157374196
np.log1p(np.exp(-3))
Out[67]: 0.04858735157374206