minor

dataloader.py

@@ -5,6 +5,7 @@
 import selfies as sf
 import yaml
 from torch.nn.utils.rnn import pad_sequence
+from pad_idx import PADDING_IDX
 
 
 def dataloader_gen(dataset_dir, percentage, vocab_path, batch_size, shuffle, drop_last=False):
@@ -23,7 +24,6 @@ def pad_collate(batch):
     """
     Put the sequences of different lengths in a minibatch by paddding.
     """
-    global PADDING_IDX
 
     lengths = [len(x) for x in batch]
 

model.py

@@ -2,12 +2,12 @@
 import torch.nn as nn
 from torch.nn.utils.rnn import pack_padded_sequence
 from torch.nn.functional import softmax
+from pad_idx import PADDING_IDX
 
 
 class RNN(torch.nn.Module):
     def __init__(self, rnn_config):
         super(RNN, self).__init__()
-        global PADDING_IDX
 
         self.embedding_layer = nn.Embedding(
             num_embeddings=rnn_config['num_embeddings'],

pad_idx.py

@@ -0,0 +1,10 @@
+# the padding index (last index) for batching
+import yaml
+
+config_dir = "./train.yaml"
+with open(config_dir, 'r') as f:
+    config = yaml.full_load(f)
+rnn_config = config['rnn_config']
+global PADDING_IDX
+PADDING_IDX = rnn_config['num_embeddings'] - 1
+print("padding index: ", PADDING_IDX)

train.py

@@ -39,13 +39,8 @@
     dataloader, train_size = dataloader_gen(
         dataset_dir, percentage, vocab_path, batch_size, shuffle, drop_last=False)
 
-    # the padding index (last index) for batching, it is set to global because
-    # the collate_fn of dataloader needs it.
-    rnn_config = config['rnn_config']
-    global PADDING_IDX
-    PADDING_IDX = rnn_config['num_embeddings'] - 1
-
     # model and training configuration
+    rnn_config = config['rnn_config']
     model = RNN(rnn_config).to(device)
     learning_rate = config['learning_rate']
     weight_decay = config['weight_decay']

train.yaml

@@ -2,7 +2,7 @@
  out_dir: '../results/run_1/'
  dataset_dir: "../zinc-smiles/"
  vocab_path: "./vocab/selfies_vocab.yaml"
- percentage: 0.2
+ percentage: 1.0
 
  rnn_config:
   # embedding
@@ -15,8 +15,8 @@
   num_layers: 3
   dropout: 0.6
 
- batch_size: 512
+ batch_size: 1024
  shuffle: True
- num_epoch: 200
+ num_epoch: 30
  learning_rate: 0.003
  weight_decay: 0.0007

vocab/selfies_vocab.yaml

@@ -0,0 +1,174 @@
+<eos>: 171
+<pad>: 173
+<sos>: 172
+'[#C-expl]': 108
+'[#C]': 116
+'[#N+expl]': 152
+'[#N]': 40
+'[#S]': 158
+'[/B-expl]': 18
+'[/B]': 26
+'[/Br]': 84
+'[/C-expl]': 7
+'[/C@@Hexpl]': 17
+'[/C@@expl]': 145
+'[/C@Hexpl]': 87
+'[/C@expl]': 27
+'[/CH-expl]': 120
+'[/C]': 29
+'[/Cexpl]': 161
+'[/Cl]': 165
+'[/F]': 22
+'[/I]': 70
+'[/N+expl]': 86
+'[/NHexpl]': 46
+'[/N]': 25
+'[/O+expl]': 90
+'[/O-expl]': 67
+'[/O]': 88
+'[/Oexpl]': 151
+'[/P@@expl]': 14
+'[/P@expl]': 4
+'[/P]': 96
+'[/S+expl]': 92
+'[/S@@expl]': 166
+'[/S@expl]': 121
+'[/S]': 80
+'[/Siexpl]': 111
+'[/Snexpl]': 68
+'[=17Oexpl]': 110
+'[=B]': 79
+'[=C]': 20
+'[=IH2expl]': 105
+'[=I]': 77
+'[=N+expl]': 125
+'[=N-expl]': 43
+'[=N]': 49
+'[=Nexpl]': 19
+'[=O+expl]': 109
+'[=O]': 170
+'[=P+expl]': 164
+'[=P@@Hexpl]': 10
+'[=P@@expl]': 107
+'[=P@Hexpl]': 1
+'[=P@expl]': 47
+'[=PHexpl]': 148
+'[=P]': 100
+'[=S+expl]': 41
+'[=S@+expl]': 48
+'[=S@@+expl]': 73
+'[=S@@expl]': 74
+'[=S@expl]': 146
+'[=SHexpl]': 150
+'[=S]': 98
+'[=Siexpl]': 157
+'[=Snexpl]': 95
+'[B-expl]': 57
+'[B@-expl]': 2
+'[B@@-expl]': 32
+'[BH-expl]': 82
+'[BH2-expl]': 39
+'[BH3-expl]': 134
+'[B]': 101
+'[Br+expl]': 94
+'[Br]': 5
+'[Branch1_1]': 123
+'[Branch1_2]': 44
+'[Branch1_3]': 35
+'[Branch2_1]': 128
+'[Branch2_2]': 56
+'[Branch2_3]': 122
+'[C+expl]': 23
+'[C-expl]': 144
+'[C@@Hexpl]': 131
+'[C@@expl]': 167
+'[C@Hexpl]': 72
+'[C@expl]': 114
+'[CH-expl]': 147
+'[CH2-expl]': 51
+'[CH2expl]': 12
+'[CHexpl]': 135
+'[C]': 140
+'[Cexpl]': 83
+'[Cl]': 65
+'[Expl/Ring1]': 91
+'[Expl/Ring2]': 138
+'[Expl=Ring1]': 30
+'[Expl=Ring2]': 154
+'[Expl\Ring1]': 133
+'[Expl\Ring2]': 50
+'[F]': 59
+'[I+expl]': 75
+'[IH2expl]': 54
+'[I]': 76
+'[N+expl]': 24
+'[N-expl]': 63
+'[N@+expl]': 169
+'[N@@+expl]': 6
+'[N@@H+expl]': 163
+'[N@H+expl]': 13
+'[NHexpl]': 97
+'[N]': 15
+'[Nexpl]': 156
+'[O+expl]': 104
+'[O-expl]': 99
+'[O]': 33
+'[Oexpl]': 153
+'[P+expl]': 160
+'[P@+expl]': 89
+'[P@@+expl]': 9
+'[P@@Hexpl]': 62
+'[P@@expl]': 21
+'[P@Hexpl]': 136
+'[P@expl]': 28
+'[PHexpl]': 3
+'[P]': 81
+'[Ring1]': 115
+'[Ring2]': 36
+'[S+expl]': 113
+'[S@+expl]': 58
+'[S@@+expl]': 139
+'[S@@Hexpl]': 130
+'[S@@expl]': 118
+'[S@Hexpl]': 53
+'[S@expl]': 168
+'[SH3expl]': 159
+'[SHexpl]': 132
+'[S]': 34
+'[Si-expl]': 117
+'[SiH3expl]': 42
+'[Siexpl]': 55
+'[Sn+2expl]': 11
+'[Sn+3expl]': 119
+'[Sn+expl]': 61
+'[SnH2+expl]': 103
+'[SnH2expl]': 142
+'[SnH4+2expl]': 8
+'[SnH6+3expl]': 69
+'[SnHexpl]': 143
+'[Snexpl]': 149
+'[\B]': 102
+'[\Br]': 124
+'[\C-expl]': 106
+'[\C@@Hexpl]': 52
+'[\C@@expl]': 137
+'[\C@Hexpl]': 129
+'[\C@expl]': 78
+'[\CH-expl]': 16
+'[\C]': 162
+'[\Cl]': 66
+'[\F]': 112
+'[\I]': 141
+'[\N+expl]': 31
+'[\NHexpl]': 127
+'[\N]': 37
+'[\O-expl]': 38
+'[\O]': 71
+'[\Oexpl]': 60
+'[\P]': 0
+'[\S+expl]': 45
+'[\S@@expl]': 93
+'[\S@expl]': 126
+'[\S]': 85
+'[\Siexpl]': 155
+'[\Snexpl]': 64