added alpha model

10 files changed, 33 insertions, 54 deletions

diff --git a/lstm_chem/__init__.py b/lstm_chem/__init__.py
index 8b13789..8b13789 100755..100644
--- a/lstm_chem/__init__.py
+++ b/lstm_chem/__init__.py
diff --git a/lstm_chem/data_loader.py b/lstm_chem/data_loader.py
index 86ddbba..243f7e6 100755..100644
--- a/lstm_chem/data_loader.py
+++ b/lstm_chem/data_loader.py
@@ -3,7 +3,7 @@ import os
 import numpy as np
 from tqdm import tqdm
 from tensorflow.keras.utils import Sequence
-from lstm_chem.utils.smiles_tokenizer import SmilesTokenizer
+from lstm_chem.utils.smiles_tokenizer2 import SmilesTokenizer
 
 
 class DataLoader(Sequence):
diff --git a/lstm_chem/finetuner.py b/lstm_chem/finetuner.py
index 904958b..24f26ce 100755..100644
--- a/lstm_chem/finetuner.py
+++ b/lstm_chem/finetuner.py
@@ -1,4 +1,4 @@
-from lstm_chem.utils.smiles_tokenizer import SmilesTokenizer
+from lstm_chem.utils.smiles_tokenizer2 import SmilesTokenizer
 from lstm_chem.generator import LSTMChemGenerator
 
 
@@ -14,7 +14,8 @@ class LSTMChemFinetuner(LSTMChemGenerator):
         self.model.compile(optimizer=self.config.optimizer,
                            loss='categorical_crossentropy')
 
-        history = self.model.fit_generator(
+#        history = self.model.fit_generator(
+        history = self.model.fit(
             self.finetune_data_loader,
             steps_per_epoch=self.finetune_data_loader.__len__(),
             epochs=self.config.finetune_epochs,
diff --git a/lstm_chem/generator.py b/lstm_chem/generator.py
index 498f864..4f80e9f 100755..100644
--- a/lstm_chem/generator.py
+++ b/lstm_chem/generator.py
@@ -1,6 +1,6 @@
 from tqdm import tqdm
 import numpy as np
-from lstm_chem.utils.smiles_tokenizer import SmilesTokenizer
+from lstm_chem.utils.smiles_tokenizer2 import SmilesTokenizer
 
 
 class LSTMChemGenerator(object):
diff --git a/lstm_chem/model.py b/lstm_chem/model.py
index 079589a..368a834 100755..100644
--- a/lstm_chem/model.py
+++ b/lstm_chem/model.py
@@ -4,7 +4,7 @@ from tensorflow.keras import Sequential
 from tensorflow.keras.models import model_from_json
 from tensorflow.keras.layers import LSTM, Dense
 from tensorflow.keras.initializers import RandomNormal
-from lstm_chem.utils.smiles_tokenizer import SmilesTokenizer
+from lstm_chem.utils.smiles_tokenizer2 import SmilesTokenizer
 
 
 class LSTMChem(object):
diff --git a/lstm_chem/trainer.py b/lstm_chem/trainer.py
index 4e8057e..0ed608a 100755..100644
--- a/lstm_chem/trainer.py
+++ b/lstm_chem/trainer.py
@@ -32,7 +32,8 @@ class LSTMChemTrainer(object):
             ))
 
     def train(self):
-        history = self.model.fit_generator(
+#        history = self.model.fit_generator(
+        history = self.model.fit(
             self.train_data_loader,
             steps_per_epoch=self.train_data_loader.__len__(),
             epochs=self.config.num_epochs,
diff --git a/lstm_chem/utils/config.py b/lstm_chem/utils/config.py
index fff7359..fff7359 100755..100644
--- a/lstm_chem/utils/config.py
+++ b/lstm_chem/utils/config.py
diff --git a/lstm_chem/utils/dirs.py b/lstm_chem/utils/dirs.py
index bcd2a49..bcd2a49 100755..100644
--- a/lstm_chem/utils/dirs.py
+++ b/lstm_chem/utils/dirs.py
diff --git a/lstm_chem/utils/smiles_tokenizer.py b/lstm_chem/utils/smiles_tokenizer.py
index d15d625..f998c4a 100755..100644
--- a/lstm_chem/utils/smiles_tokenizer.py
+++ b/lstm_chem/utils/smiles_tokenizer.py
@@ -1,8 +1,5 @@
-import copy
 import numpy as np
 
-import time
-
 
 class SmilesTokenizer(object):
     def __init__(self):
@@ -30,7 +27,7 @@ class SmilesTokenizer(object):
             'S',
             'K',
             'V',
-            'I',
+            'I'
         ]
         special = [
             '(', ')', '[', ']', '=', '#', '%', '0', '1', '2', '3', '4', '5',
@@ -51,8 +48,6 @@ class SmilesTokenizer(object):
         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]
@@ -60,8 +55,6 @@ class SmilesTokenizer(object):
                     token.append(symbol)
                     i += len(symbol)
                     break
-            if time.time() > timeout:
-                break 
         return token
 
     def one_hot_encode(self, tokenized_smiles):
diff --git a/lstm_chem/utils/smiles_tokenizer.py.4faeffb638548d04ca4415dfe32cf8c7.py b/lstm_chem/utils/smiles_tokenizer2.py
index d15d625..29575ba 100755..100644
--- a/lstm_chem/utils/smiles_tokenizer.py.4faeffb638548d04ca4415dfe32cf8c7.py
+++ b/lstm_chem/utils/smiles_tokenizer2.py
@@ -1,36 +1,11 @@
-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',
+            'Al', 'As', 'B', 'Br', 'C', 'Cl', 'F', 'H', 'I', 'K', 'Li', 'N',
+            'Na', 'O', 'P', 'S', 'Se', 'Si', 'Te'
         ]
         special = [
             '(', ')', '[', ']', '=', '#', '%', '0', '1', '2', '3', '4', '5',
@@ -39,29 +14,38 @@ class SmilesTokenizer(object):
         padding = ['G', 'A', 'E']
 
         self.table = sorted(atoms, key=len, reverse=True) + special + padding
-        self.table_len = len(self.table)
+        table_len = len(self.table)
+
+        self.table_2_chars = list(filter(lambda x: len(x) == 2, self.table))
+        self.table_1_chars = list(filter(lambda x: len(x) == 1, self.table))
 
         self.one_hot_dict = {}
         for i, symbol in enumerate(self.table):
-            vec = np.zeros(self.table_len, dtype=np.float32)
+            vec = np.zeros(table_len, dtype=np.float32)
             vec[i] = 1
             self.one_hot_dict[symbol] = vec
 
     def tokenize(self, smiles):
+        smiles = smiles + ' '
         N = len(smiles)
-        i = 0
         token = []
-
-        timeout = time.time() + 5   # 5 seconds from now
+        i = 0
         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 
+            c1 = smiles[i]
+            c2 = smiles[i:i + 2]
+
+            if c2 in self.table_2_chars:
+                token.append(c2)
+                i += 2
+                continue
+
+            if c1 in self.table_1_chars:
+                token.append(c1)
+                i += 1
+                continue
+
+            i += 1
+
         return token
 
     def one_hot_encode(self, tokenized_smiles):