diff options
author | Alex Auvolat <alex@adnab.me> | 2017-01-17 16:35:28 +0100 |
---|---|---|
committer | Alex Auvolat <alex@adnab.me> | 2017-01-17 16:35:28 +0100 |
commit | 2a4f89c41b82e7173d6fb4481af0dcd58c812c19 (patch) | |
tree | d99f1cf430c86b54449de1ae1fb66dfd00f9c098 /config/lstm-deep-l1.py | |
parent | 9524ab7589f1083608cae5608e16285b0707bc48 (diff) | |
download | text-rnn-2a4f89c41b82e7173d6fb4481af0dcd58c812c19.tar.gz text-rnn-2a4f89c41b82e7173d6fb4481af0dcd58c812c19.zip |
Stuff.
Diffstat (limited to 'config/lstm-deep-l1.py')
-rw-r--r-- | config/lstm-deep-l1.py | 52 |
1 files changed, 52 insertions, 0 deletions
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' + |