Mercurial > pylearn
comparison doc/v2_planning/architecture.txt @ 1186:f111f8c2a280
Update after sept. 17th meeting
| author | Pascal Lamblin <lamblinp@iro.umontreal.ca> |
|---|---|
| date | Fri, 17 Sep 2010 17:07:52 -0400 |
| parents | ae5ba6206fd3 |
| children | 9ff2242a817b |
comparison
equal
deleted
inserted
replaced
| 1185:4ea46ef9822a | 1186:f111f8c2a280 |
|---|---|
| 138 come with constraints on what fn, args, and kwargs can be. Specifically, they | 138 come with constraints on what fn, args, and kwargs can be. Specifically, they |
| 139 must be picklable, and there are benefits (e.g. automatic function call caching) | 139 must be picklable, and there are benefits (e.g. automatic function call caching) |
| 140 associated with them being hashable as well. | 140 associated with them being hashable as well. |
| 141 | 141 |
| 142 | 142 |
| 143 Benchmark | |
| 144 ========= | |
| 143 | 145 |
| 146 During the general meeting on sept. 17th, we agreed to produce at least pseudo-code (if | |
| 147 possible, actual code) for the following model: | |
| 148 A Deep Belief Net (with greedy layerwise pre-training, and supervised | |
| 149 fine-tuning), with preprocessing of the data, double cross-validation, and | |
| 150 save/load of the model. | |
| 151 | |
| 152 The different approach to be tested are: | |
| 153 - Plugins with a global scheduler driving the experiment (Razvan's team) | |
| 154 - Objects, with basic hooks at predefined places (Pascal L.'s team) | |
| 155 - Existing objects and code (including dbi and Jobman), with some more | |
| 156 pieces to tie things together (Fred B.) | |
| 157 |