-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathMCTSmain.py
More file actions
200 lines (177 loc) · 10.2 KB
/
MCTSmain.py
File metadata and controls
200 lines (177 loc) · 10.2 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
from copy import deepcopy
import math
from mol_preprocessing_prediction import *
def return_solutions(new_node_number):
"""
function saves list to rdf file with synthesis ways of target molecule
"""
with RDFwrite("/home/aigul/Retro/templates/first_predictions.rdf") as f:
global synthetic_path
parent_node_number = lemon_tree.go_to_parent(new_node_number)
if new_node_number != parent_node_number:
synthetic_path.append(
lemon_tree.return_reaction(parent_node_number=parent_node_number, new_node_number=new_node_number))
return_solutions(parent_node_number)
for solution in synthetic_path:
f.write(solution)
return synthetic_path
synthetic_path = []
def update(new_node_number, reward):
"""
function updates Q, a, number of visits and reward of node, works recursive until node not root of tree
"""
node_attrs_2 = {"number_of_visits": lemon_tree.nodes[new_node_number]["number_of_visits"] + 1,
"reward": lemon_tree.nodes[new_node_number]["reward"] + (reward)}
lemon_tree.change_atrrs_of_node(node_number=new_node_number, dict_with_attrs=node_attrs_2)
Q = (1 / (lemon_tree.nodes[new_node_number]["number_of_visits"])) * (lemon_tree.nodes[new_node_number]["reward"])
P = lemon_tree.nodes[new_node_number]["probability"] # prob from NN
parent_node = lemon_tree.go_to_parent(new_node_number)
N = lemon_tree.nodes[new_node_number]["number_of_visits"]
N_pred = lemon_tree.nodes[parent_node]["number_of_visits"]
a = (Q) + P * (math.sqrt(N_pred) / (1 + N))
node_attrs_3 = {"Q": Q, "a": a}
lemon_tree.change_atrrs_of_node(node_number=new_node_number, dict_with_attrs=node_attrs_3)
if parent_node != 1:
update(parent_node, reward)
def expansion(node_number):
"""
function adds nodes with predicted reactants. if nothing is predicted starts updating node with -1 reward
:param node_number:
:return: list of added nodes
"""
global solution_found_counter
max_depth = 10
current_node = lemon_tree.go_to_node(node_number)
list_of_reagents = current_node["reagents"]
if list_of_reagents == []:
lemon_tree.change_atrrs_of_node(node_number=node_number, dict_with_attrs={"expanded": True})
return
mol_container = list_of_reagents[0]
lemon_tree.change_atrrs_of_node(node_number=node_number, dict_with_attrs={"expanded": True})
new_patterns = return_patterns_expansion(prediction(prep_mol_for_nn(mol_container)))
if len(new_patterns) == 0:
lemon_tree.nodes[node_number]["terminal"] = True
update(node_number, -1)
if lemon_tree.nodes[node_number]["terminal"] != True:
new_mols_from_pred = create_mol_from_pattern(new_patterns, mol_container)
for new_mol in new_mols_from_pred:
if len(new_mol[0]) == 0:
lemon_tree.change_atrrs_of_node(node_number=node_number, dict_with_attrs={"terminal": True})
update(node_number, -1)
else:
for j2 in range(len(new_mol[0])):
copy_of_list_of_reagents = deepcopy(list_of_reagents[1:])
# check compounds in DB and if yes exclude it from node molecule list
for j3 in new_mol[0][j2]:
if j3.get_signature_hash() not in reagents_in_store:
copy_of_list_of_reagents.append(j3)
# check if node exist if Rollout = False? if not:
# in rollout == False we add new nodes, 1) if children of nodes absent
# and 2) if list of hashes != list of new list of hashes; if we have one prediction and this prediction in node we continue
if lemon_tree.go_to_child(node_number) == []:
new_node_number = lemon_tree.add_node_(list_of_reagents=copy_of_list_of_reagents,
parent_node_number=node_number,
probability=new_mol[2][j2])
lemon_tree.add_edge_(parent_node_number=node_number, node_number=new_node_number,
reaction=new_mol[1][j2])
else:
repeated = False
for child_num in range(len(lemon_tree.go_to_child(node_number))):
if lemon_tree.nodes[lemon_tree.go_to_child(node_number)[child_num]][
"reagents"] == copy_of_list_of_reagents:
repeated = True
if repeated == False:
new_node_number = lemon_tree.add_node_(list_of_reagents=copy_of_list_of_reagents,
parent_node_number=node_number,
probability=new_mol[2][j2])
lemon_tree.add_edge_(parent_node_number=node_number, node_number=new_node_number,
reaction=new_mol[1][j2])
else:
lemon_tree.change_atrrs_of_node(node_number=node_number, dict_with_attrs={"expanded": True})
return lemon_tree.go_to_child(node_number)
if lemon_tree.node_depth(new_node_number) > max_depth and lemon_tree.node_solved(
new_node_number) is False:
update(new_node_number, -1)
elif len(lemon_tree.nodes[new_node_number]["reagents"]) == 0:
lemon_tree.change_atrrs_of_node(node_number=new_node_number,
dict_with_attrs={"solved": True, "expanded": True,
"terminal": True})
solution_found_counter += 1
return_solutions(new_node_number)
update(new_node_number, 1)
return lemon_tree.go_to_child(parent_node_number=node_number)
def rollout(node_number):
"""
function adds nodes with predicted reactants. if nothing is predicted starts updating node with -1 reward, if node
not terminal recursively starts rollout and ends when model can't predict nothing, or if node has max depth, or
molecule in list of reagents in store
:param node_number:
:return:
"""
global solution_found_counter
max_depth = 10
current_node = lemon_tree.go_to_node(node_number)
list_of_reagents = current_node["reagents"]
if list_of_reagents == []:
return
mol_container = list_of_reagents[0]
new_patterns = return_patterns_rollout(prediction(prep_mol_for_nn(mol_container)))
if len(new_patterns) == 0:
lemon_tree.nodes[node_number]["terminal"] = True
update(node_number, -1)
if lemon_tree.nodes[node_number]["terminal"] != True:
new_mols_from_pred = create_mol_from_pattern(new_patterns, mol_container)
for new_mol in new_mols_from_pred:
if len(new_mol[0]) == 0:
lemon_tree.change_atrrs_of_node(node_number=node_number, dict_with_attrs={"terminal": True})
update(node_number, -1)
else:
for j2 in range(len(new_mol[0])):
copy_of_list_of_reagents = deepcopy(list_of_reagents[1:])
# check compounds in DB and if yes exclude it from node molecule list
for j3 in new_mol[0][j2]:
if j3.get_signature_hash() not in reagents_in_store:
copy_of_list_of_reagents.append(j3)
# check if node exist if Rollout = False? if not:
# in rollout == False we add new nodes, 1) if children of nodes absent
# and 2) if list of hashes != list of new list of hashes; if we have one prediction and this prediction in node we continue
if lemon_tree.go_to_child(node_number) == []:
new_node_number = lemon_tree.add_node_(list_of_reagents=copy_of_list_of_reagents,
parent_node_number=node_number,
probability=new_mol[2][j2])
lemon_tree.add_edge_(parent_node_number=node_number, node_number=new_node_number,
reaction=new_mol[1][j2])
else:
break
if lemon_tree.node_depth(new_node_number) > max_depth and lemon_tree.node_solved(
new_node_number) is False:
update(new_node_number, -1)
elif len(lemon_tree.nodes[new_node_number]["reagents"]) == 0:
lemon_tree.change_atrrs_of_node(node_number=new_node_number,
dict_with_attrs={"solved": True,
"terminal": True})
solution_found_counter += 1
return_solutions(new_node_number)
update(new_node_number, 1)
else:
node_nums_rollout.append(new_node_number)
rollout(new_node_number)
def MCTsearch(Max_Iteration, Max_Num_Solved):
max_depth = 10
for i in range(Max_Iteration):
if solution_found_counter < Max_Num_Solved:
node_number = 1
new_node_number = lemon_tree.search(node_number) or lemon_tree.random_search(node_number)
if not new_node_number:
continue
if lemon_tree.node_depth(new_node_number) < max_depth:
expanded_nodes_nums = expansion(new_node_number)
if expanded_nodes_nums:
for expanded_node in expanded_nodes_nums:
rollout(expanded_node)
else:
continue
synthetic_path = []
solution_found_counter = 0
node_nums_rollout = []
MCTsearch(Max_Iteration=100, Max_Num_Solved=2)