Mercurial > ift6266
comparison deep/stacked_dae/nist_sda.py @ 186:d364a130b221
Ajout du code de base pour scalar_series. Modifications à stacked_dae: réglé un problème avec les input_divider (empêchait une optimisation), et ajouté utilisation des séries. Si j'avais pas déjà commité, aussi, j'ai enlevé l'histoire de réutilisation du pretraining: c'était compliqué (error prone) et ça créait des jobs beaucoup trop longues.
author | fsavard |
---|---|
date | Mon, 01 Mar 2010 11:45:25 -0500 |
parents | b9ea8e2d071a |
children | 3632e6258642 c69c1d832a53 |
comparison
equal
deleted
inserted
replaced
185:b9ea8e2d071a | 186:d364a130b221 |
---|---|
30 from scalar_series import * | 30 from scalar_series import * |
31 SERIES_AVAILABLE = True | 31 SERIES_AVAILABLE = True |
32 except ImportError: | 32 except ImportError: |
33 print "Could not import Series" | 33 print "Could not import Series" |
34 | 34 |
35 TEST_CONFIG = True | 35 TEST_CONFIG = False |
36 | 36 |
37 NIST_ALL_LOCATION = '/data/lisa/data/nist/by_class/all' | 37 NIST_ALL_LOCATION = '/data/lisa/data/nist/by_class/all' |
38 | 38 |
39 JOBDB = 'postgres://ift6266h10@gershwin/ift6266h10_db/fsavard_sda2' | 39 JOBDB = 'postgres://ift6266h10@gershwin/ift6266h10_db/fsavard_sda2' |
40 | 40 |
44 if TEST_CONFIG: | 44 if TEST_CONFIG: |
45 REDUCE_TRAIN_TO = 1000 | 45 REDUCE_TRAIN_TO = 1000 |
46 MAX_FINETUNING_EPOCHS = 2 | 46 MAX_FINETUNING_EPOCHS = 2 |
47 REDUCE_EVERY = 10 | 47 REDUCE_EVERY = 10 |
48 | 48 |
49 EXPERIMENT_PATH = "ift6266.scripts.stacked_dae.nist_sda.jobman_entrypoint" | 49 EXPERIMENT_PATH = "ift6266.deep.stacked_dae.nist_sda.jobman_entrypoint" |
50 | 50 |
51 JOB_VALS = {'pretraining_lr': [0.1, 0.01],#, 0.001],#, 0.0001], | 51 JOB_VALS = {'pretraining_lr': [0.1, 0.01],#, 0.001],#, 0.0001], |
52 'pretraining_epochs_per_layer': [10,20], | 52 'pretraining_epochs_per_layer': [10,20], |
53 'hidden_layers_sizes': [300,800], | 53 'hidden_layers_sizes': [300,800], |
54 'corruption_levels': [0.1,0.2,0.3], | 54 'corruption_levels': [0.1,0.2,0.3], |