added Curie-Generate BETA

1 files changed, 72 insertions, 0 deletions

diff --git a/lstm_chem/utils/smiles_tokenizer.py b/lstm_chem/utils/smiles_tokenizer.py
new file mode 100755
index 0000000..d15d625
--- /dev/null
+++ b/lstm_chem/utils/smiles_tokenizer.py
@@ -0,0 +1,72 @@
+import copy
+import numpy as np
+
+import time
+
+
+class SmilesTokenizer(object):
+    def __init__(self):
+        atoms = [
+            'Li',
+            'Na',
+            'Al',
+            'Si',
+            'Cl',
+            'Sc',
+            'Zn',
+            'As',
+            'Se',
+            'Br',
+            'Sn',
+            'Te',
+            'Cn',
+            'H',
+            'B',
+            'C',
+            'N',
+            'O',
+            'F',
+            'P',
+            'S',
+            'K',
+            'V',
+            'I',
+        ]
+        special = [
+            '(', ')', '[', ']', '=', '#', '%', '0', '1', '2', '3', '4', '5',
+            '6', '7', '8', '9', '+', '-', 'se', 'te', 'c', 'n', 'o', 's'
+        ]
+        padding = ['G', 'A', 'E']
+
+        self.table = sorted(atoms, key=len, reverse=True) + special + padding
+        self.table_len = len(self.table)
+
+        self.one_hot_dict = {}
+        for i, symbol in enumerate(self.table):
+            vec = np.zeros(self.table_len, dtype=np.float32)
+            vec[i] = 1
+            self.one_hot_dict[symbol] = vec
+
+    def tokenize(self, smiles):
+        N = len(smiles)
+        i = 0
+        token = []
+
+        timeout = time.time() + 5   # 5 seconds from now
+        while (i < N):
+            for j in range(self.table_len):
+                symbol = self.table[j]
+                if symbol == smiles[i:i + len(symbol)]:
+                    token.append(symbol)
+                    i += len(symbol)
+                    break
+            if time.time() > timeout:
+                break 
+        return token
+
+    def one_hot_encode(self, tokenized_smiles):
+        result = np.array(
+            [self.one_hot_dict[symbol] for symbol in tokenized_smiles],
+            dtype=np.float32)
+        result = result.reshape(1, result.shape[0], result.shape[1])
+        return result