Mercurial > pylearn
comparison doc/v2_planning/requirements.txt @ 1243:7b86c1772932
requirements: Agreement with Pascal's remark
| author | Olivier Delalleau <delallea@iro> |
|---|---|
| date | Thu, 23 Sep 2010 12:10:41 -0400 |
| parents | 6801451a86bb |
| children | e5b7a7913329 |
comparison
equal
deleted
inserted
replaced
| 1241:6801451a86bb | 1243:7b86c1772932 |
|---|---|
| 114 | 114 |
| 115 PL: I'm not sure it the job should have to return its current state. I think | 115 PL: I'm not sure it the job should have to return its current state. I think |
| 116 it would be enough that it returns a consistent checkpoint, even if it is from | 116 it would be enough that it returns a consistent checkpoint, even if it is from |
| 117 some time in the past (ideally, not more than a few minutes ago). | 117 some time in the past (ideally, not more than a few minutes ago). |
| 118 | 118 |
| 119 OD: I agree with PL. | |
| 120 | |
| 119 OD asks: When you say "The experiment cannot assume that resources written by | 121 OD asks: When you say "The experiment cannot assume that resources written by |
| 120 the experiment will still be there", do you mean we should be able to recover | 122 the experiment will still be there", do you mean we should be able to recover |
| 121 the exact same output after interrupting an experiment, wiping its expdir, and | 123 the exact same output after interrupting an experiment, wiping its expdir, and |
| 122 restarting it? This would mean that any output saved on disk by the experiment | 124 restarting it? This would mean that any output saved on disk by the experiment |
| 123 also has to be serialized within the experiment, which may lead to very big | 125 also has to be serialized within the experiment, which may lead to very big |