adding code and data

config.py

@@ -0,0 +1,53 @@
+###preprocessing options###
+word_minfreq=3
+
+###hyper-parameters###
+seed=0
+batch_size=32
+keep_prob=0.7
+epoch_size=30
+max_grad_norm=5
+#language model
+word_embedding_dim=100
+word_embedding_model="pretrain_word2vec/dim100/word2vec.bin"
+lm_enc_dim=200
+lm_dec_dim=600
+lm_dec_layer_size=1
+lm_attend_dim=25
+lm_learning_rate=0.2
+#pentameter model
+char_embedding_dim=150
+pm_enc_dim=50
+pm_dec_dim=200
+pm_attend_dim=50
+pm_learning_rate=0.001
+repeat_loss_scale=1.0
+cov_loss_scale=1.0
+cov_loss_threshold=0.7
+sigma=1.00
+#rhyme model
+rm_dim=100
+rm_neg=5 #extra randomly sampled negative examples
+rm_delta=0.5
+rm_learning_rate=0.001
+
+###sonnet hyper-parameters###
+bptt_truncate=2 #number of sonnet lines to truncate bptt
+doc_lines=14 #total number of lines for a sonnet
+
+###misc###
+verbose=False
+save_model=True
+
+###input/output###
+output_dir="output_tmp"
+train_data="datasets/gutenberg/sonnet_train.txt"
+valid_data="datasets/gutenberg/sonnet_valid.txt"
+test_data="datasets/gutenberg/sonnet_test.txt"
+pentameter_pos=None #"datasets/gutenberg/pentameter_pos.txt"
+pentameter_neg=None #"datasets/gutenberg/pentameter_neg.txt"
+output_prefix="wmin%d_sd%d_bat%d_kp%.1f_eph%d_grd%d_wdim%d_lmedim%d_lmddim%d_lmdlayer%d_lmadim%d_lmlr%.1f_cdim%d_pmedim%d_pmddim%d_pmadim%d_pmlr%.1E_loss%.1f-%.1f-%.1f_sm%.2f_rmdim%d_rmn%d_rmd%.1f_rmlr%.1E_son%d-%d" % \
+    (word_minfreq, seed, batch_size, keep_prob, epoch_size, max_grad_norm, word_embedding_dim, lm_enc_dim,
+    lm_dec_dim, lm_dec_layer_size, lm_attend_dim, lm_learning_rate,
+    char_embedding_dim, pm_enc_dim, pm_dec_dim, pm_attend_dim, pm_learning_rate, repeat_loss_scale,
+    cov_loss_scale, cov_loss_threshold, sigma, rm_dim, rm_neg, rm_delta, rm_learning_rate, bptt_truncate, doc_lines)

pretrain_word2vec.py

@@ -0,0 +1,22 @@
+import gensim.models as g
+import logging
+import os
+
+#parameters
+output_dir="pretrain_word2vec/"
+input_doc="datasets/gutenberg/pretrain.txt"
+
+#main
+if not os.path.exists(output_dir):
+    os.makedirs(output_dir)
+
+#logging
+logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.INFO)
+
+docs = g.word2vec.LineSentence(input_doc)
+m = g.Word2Vec(docs, size=100, alpha=0.025, window=5, min_count=3, \
+    sample=1e-5, workers=16, min_alpha=0.0001, sg=1, hs=0, negative=5, iter=200)
+
+#save model
+m.save(output_dir + "/word2vec.bin")
+m.wv.save_word2vec_format(output_dir + "/word2vec.txt", binary=False)