diff options
Diffstat (limited to 'config')
| -rw-r--r-- | config/condlstm.py | 85 | ||||
| -rw-r--r-- | config/lstm-deep-l1.py | 52 | ||||
| -rw-r--r-- | config/lstm-frigo-irc.py | 1 |
3 files changed, 138 insertions, 0 deletions
diff --git a/config/condlstm.py b/config/condlstm.py new file mode 100644 index 0000000..f846972 --- /dev/null +++ b/config/condlstm.py @@ -0,0 +1,85 @@ +from blocks.algorithms import AdaDelta +from blocks.bricks import Tanh +from blocks.initialization import IsotropicGaussian, Constant + +from model.condlstm import Model + +dataset = 'data/logcompil.txt' +io_dim = 256 + +# An epoch will be composed of 'num_seqs' sequences of len 'seq_len' +# divided in chunks of lengh 'seq_div_size' +num_seqs = 50 +seq_len = 5000 +seq_div_size = 200 + +layers = [ + # Slowlier + {'dim': 128, + 'reset_after': ' \t\n,.:;/!?()[]{}<>\\\'"*+-^_|#~&`@$%', + }, + {'dim': 128, + 'run_on': ' \t\n,.:;/!?()[]{}<>\\\'"*+-^_|#~&`@$%', + 'reset_after': ' \t\n', + }, + {'dim': 128, + 'run_on': ' \t\n', + 'reset_after': '\t\n', + }, + {'dim': 256, + 'run_on': ' \t\n', + 'reset_after': '\n', + }, + {'dim': 512, + 'run_on': '\t\n', + }, + # Slowest + {'dim': 512, + 'run_on': '\n', + }, + # Fastlier + {'dim': 512, + 'run_on': '\t\n', + }, + {'dim': 256, + 'run_on': ' \t\n', + 'reset_before':'\n', + }, + {'dim': 128, + 'run_on': ' \t\n', + 'reset_before': '\t\n', + }, + {'dim': 128, + 'run_on': ' \t\n,.:;/!?()[]{}<>\\\'"*+-^_|#~&`@$%', + 'reset_before': ' \t\n', + }, + {'dim': 128, + 'reset_before': ' \t\n,.:;/!?()[]{}<>\\\'"*+-^_|#~&`@$%', + }, +] +activation_function = Tanh() + +w_noise_std = 0 +i_dropout = 0 + +l1_reg = 0 +l1_reg_weight = 0 + +step_rule = AdaDelta() + +weights_init = IsotropicGaussian(0.01) +biases_init = Constant(0.) + +# parameter saving freq (number of batches) +monitor_freq = 100 +save_freq = 100 + +# used for sample generation and IRC mode +#sample_temperature = 0.7 #0.5 +sample_temperature = 0.9 #0.5 + +# do we want to generate samples at times during training? +sample_len = 500 +sample_freq = 100 +sample_init = '\nalex\ttu crois?\n' + diff --git a/config/lstm-deep-l1.py b/config/lstm-deep-l1.py new file mode 100644 index 0000000..fd330c7 --- /dev/null +++ b/config/lstm-deep-l1.py @@ -0,0 +1,52 @@ +from blocks.algorithms import AdaDelta +from blocks.bricks import Tanh + +from model.lstm import Model + +dataset = 'data/logcompil.txt' +io_dim = 256 + +# An epoch will be composed of 'num_seqs' sequences of len 'seq_len' +# divided in chunks of lengh 'seq_div_size' +num_seqs = 100 +seq_len = 5000 +seq_div_size = 200 + +layers = [ + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, + {'dim': 512}, +] +activation_function = Tanh() + +i2h_all = True # input to all hidden layers or only first layer +h2o_all = True # all hiden layers to output or only last layer + +w_noise_std = 0 +i_dropout = 0 + +l1_reg = 0 +l1_reg_weight = 0.05 + +step_rule = AdaDelta() + +# parameter saving freq (number of batches) +monitor_freq = 100 +save_freq = 100 + +# used for sample generation and IRC mode +#sample_temperature = 0.7 #0.5 +sample_temperature = 0.9 #0.5 + +# do we want to generate samples at times during training? +sample_len = 1000 +sample_freq = 100 +sample_init = '\nalex\ttu crois?\n' + diff --git a/config/lstm-frigo-irc.py b/config/lstm-frigo-irc.py index 04468c9..38f0766 100644 --- a/config/lstm-frigo-irc.py +++ b/config/lstm-frigo-irc.py @@ -26,6 +26,7 @@ w_noise_std = 0.02 i_dropout = 0.5 l1_reg = 0 +l1_reg_weight = 0 step_rule = AdaDelta() |