I'd note that `Vector.sort()` is an awkward choice of name for a function that returns a result, rather than mutating the object in-place, given the standard library tends to use infinitive verbs to mean in place functions, while past tense verbs return a new value with the modification applied.
e.g: `list.sort()` is in-place, `sorted(list)` returns a new list.
True, but who really wants the mutating version? Worse, do we really want to make it more obvious/natural to modify in place? I think this is a standard library convention that is best forgotten. In pandas, for example, you need to write inplace=True to use the mutating versions of methods, which hopefully makes mutation awkward enough to discourage it.
More often than not, I want the in-place version. When dealing with a significant amount of data, it's better not to make unnecessary copies. What is your argument for wanting to discourage mutation? Either way, the argument of the parent comment is clear, it's best to follow established naming conventions.
Agree. Naming is difficult and here maybe I did a wrong choice. Anyway, pykov.Vector class inherit from dict, which is not ordered, so mutating in place cannot be done. The sort() method returns a list, which makes the name "sort" even more problematic. What do you suggest? (OrderedDict?)
I coded this library during my PhD, where I applied Markov Chains to the study of protein folding dynamics. I am quite confident with it, but as far as I know few others used it. At that time, I used pykov with networks of around 100k nodes and 500k links. Only recently I replaced pysparse with scipy sparse solvers. Entire methods are comment out because they are heuristics that do not belong to the standard Markov Theory, and they can be removed.
1. Separate python 2 and python 3 modules? How can I be sure that they both do the same thing?
2. You have the tarball in your source control. Why?
3. You have a 1.7k line file. Something tells me that they can be split up into smaller modules.
4. Commented out code? Why is it in there? If it's unused, remove it.
5. No lines between methods, single line between classes. Please have a look at PEP8.
6. You have that .pysparse file in source control. Why?
7. You say that Pyvok depends on Scipy and Numpy but you don't specify the dependency in your setup.py file. Anyone that installs the package from PyPI is going to have a broken package.
Pykov implements the computation of some of the most common quantities related to discrete-time finite regular Markov Chains, namely: steady state, mean first passage times and absorbing times. The calculation are performed by means of the analytic formulas described on Kemeny&Snell, and the steady state is derived with the inverse iteration method. If you are interested on random walkers, pykov offers a handy way to generate them and evaluate their probability.
Sorry, I do not know PyMC well enough to say if the above quantities can be calculated also with it.
13 comments
[ 2.2 ms ] story [ 39.1 ms ] threadI want miracles, not retard A.I.
e.g: `list.sort()` is in-place, `sorted(list)` returns a new list.
The main reason is that mutable objects are more complex and thus harder to reason about (which leads to more bugs).
Reading the source sets a couple of alarm bells ringing -
entirely separate and independent py2k py3k versions?
entire methods just commented out?
and as latty points out wandering from established naming conventions...
1. Separate python 2 and python 3 modules? How can I be sure that they both do the same thing?
2. You have the tarball in your source control. Why?
3. You have a 1.7k line file. Something tells me that they can be split up into smaller modules.
4. Commented out code? Why is it in there? If it's unused, remove it.
5. No lines between methods, single line between classes. Please have a look at PEP8.
6. You have that .pysparse file in source control. Why?
7. You say that Pyvok depends on Scipy and Numpy but you don't specify the dependency in your setup.py file. Anyone that installs the package from PyPI is going to have a broken package.
I would be interested in a Python Markov chain library since I use them a lot.