| 1 |
What follows is an old TODO. Nothing of what you read here is going to ever be implemented, |
| 2 |
unless some new developer shows up. We keep the TODO here just for historical reasons and |
| 3 |
for everlasting shame on us ;) |
| 4 |
|
| 5 |
for MDP 3.3 |
| 6 |
=========== |
| 7 |
|
| 8 |
- clearify documentation on Flow.train as suggested by Fabian Schoenfeld in |
| 9 |
http://sourceforge.net/mailarchive/message.php?msg_id=27926167 |
| 10 |
- checkout windows 64bit binaries from http://www.lfd.uci.edu/~gohlke/pythonlibs/ maybe point to it from the documentation? |
| 11 |
- add code for generating the plots in the tutorial (but we don't want the |
| 12 |
doctests to fail if matplotlib is not installed: find a good way of |
| 13 |
dealing with it, pymvpa already does it properly) |
| 14 |
- document extension disable environment variables |
| 15 |
- docstrings should be migrated to rst everywhere (only class.__doc__ has |
| 16 |
been done for now) |
| 17 |
- example pages are still not good enough: |
| 18 |
idea: I actually think that every example page should be |
| 19 |
self-consistent, so no imports (apart from system-installed modules |
| 20 |
of course) should be allowed. if an example requires a lot of |
| 21 |
boiler-plate code, this code could be attached at the end of the |
| 22 |
example page. note that you can use the literalinclude [1] |
| 23 |
directive to include the code, so you don't even need to re-type |
| 24 |
everything in rst. this solution has the big advantage of allowing |
| 25 |
doctest to run properly and to avoid magic tricks with sys.path, |
| 26 |
which are not exactly elegant and prone to fail |
| 27 |
[1] http://sphinx.pocoo.org/markup/code.html?highlight=include#literalinclude |
| 28 |
- remove _monkeypatch_pp when parallel python is fixed |
| 29 |
- let EtaConmputerNode match the new convention of SFA Node in terms of |
| 30 |
last sample. |
| 31 |
- add example of usage of MDP within PyMVPA. The code exists already: |
| 32 |
https://github.com/PyMVPA/PyMVPA/blob/master/doc/examples/mdp_mnist.py |
| 33 |
- find a solution to the __revision__ problem: should it be set on installation? |
| 34 |
numpy solved the problem, do we want to go this route? |
| 35 |
- parallel: provide adapters for more sophisticated schedulers |
| 36 |
- add _bias attribute to PCANode to make it more consistent with SFA node. |
| 37 |
Maybe one could even create a new AffineNode node as a base class for PCA, |
| 38 |
SFA and other affine transformations? |
| 39 |
This might also be a good opportunity for some more PEP8 cleanup. |
| 40 |
- add more classifier stuff, like a ClassifierFlow |
| 41 |
- add an example of training a node with multiple training phases using a flow, |
| 42 |
where the training is done first using lists and then using a custom iterators. |
| 43 |
special care should be taken on explaining how to pass additional arguments to |
| 44 |
the train method. an example of how this can look confusing and go wrong can be |
| 45 |
found in the testFlowWrongItarableException test in test_flows.py |
| 46 |
- fix damned LLENode test for 2D shape embedded in 3D! |
| 47 |
- check that SparsePCA works on machine with scipy 0.9.0, add it to MDP if so |
| 48 |
- create a Flow metaclass to enable Flow extensions (think of ParallelFlow!) |
| 49 |
- implement an extension context manager with additional parameters and |
| 50 |
exception handling. E.g.: |
| 51 |
flow = Flow([PCANode()]) |
| 52 |
with extension('parallel', args=(ProcessScheduler,8)): |
| 53 |
flow.train(x) |
| 54 |
note that the context manager takes care of initialize and shutting down the |
| 55 |
scheduler. Proposed syntax: |
| 56 |
extension(string, args=tuple, kwargs=dictionary) |
| 57 |
- bimdp: add deep belief network flow and nodes to the core bimdp |
| 58 |
- add cross-correlation tools, maybe support the use of a parallel scheduler |
| 59 |
- check problem with LLENode tutorial demo when using matplotlib |
| 60 |
0.99.1.2, see Olivier Grisel's email |
| 61 |
- LinearRegressionNode: add optional 2nd phase that computes residuals |
| 62 |
and significance of the slope |
| 63 |
- provide a Node pickler, for arrays use the binary numpy format (npy, |
| 64 |
numpy.save, numpy.load) and not pickle: pickling arrays is unsupported |
| 65 |
- add benchmarks for parallel module |
| 66 |
- provide different versions of the MDP logo which includes the |
| 67 |
website address, possibly one higher quality print version, |
| 68 |
available in "how to cite" section |
| 69 |
- Use the new property decorators when migrating to Python 2.6 (see |
| 70 |
http://docs.python.org/library/functions.html#property). |
| 71 |
- kalman filters |
| 72 |
- memory profiler |
| 73 |
- GUI |