def train_ex():
agent.load_state_dict(torch.load(params['pr_path'] + '.pkg', map_location=utils.dev))
prior = GraphModel(voc)
prior.load_state_dict(torch.load(params['ft_path'] + '.pkg', map_location=utils.dev))
evolver = GraphExplorer(agent, mutate=prior)
evolver.batch_size = BATCH_SIZE
evolver.epsilon = float(OPT.get('-e', '1e-2'))
evolver.sigma = float(OPT.get('-b', '0.00'))
evolver.scheme = OPT.get('-s', 'WS')
evolver.repeat = 1
keys = ['A2A', 'QED']
A2A = utils.Predictor('output/env/RF_%s_CHEMBL251.pkg' % z, type=z)
QED = utils.Property('QED')
# Chose the desirability function
objs = [A2A, QED]
if evolver.scheme == 'WS':
mod1 = utils.ClippedScore(lower_x=3, upper_x=10)
mod2 = utils.ClippedScore(lower_x=0, upper_x=1.0)
ths = [0.5, 0]
else:
mod1 = utils.ClippedScore(lower_x=3, upper_x=6.5)
mod2 = utils.ClippedScore(lower_x=0, upper_x=1.0)
ths = [0.99, 0]
mods = [mod1, mod2]
evolver.env = utils.Env(objs=objs, mods=mods, keys=keys, ths=ths)
# import evolve as agent
evolver.out = root + '/%s_%s_%.0e' % (alg, evolver.scheme, evolver.epsilon)
evolver.fit(train_loader, test_loader=valid_loader)
def rl_train():
opts, args = getopt.getopt(sys.argv[1:], "a:e:b:g:c:s:z:")
OPT = dict(opts)
case = OPT['-c'] if '-c' in OPT else 'OBJ1'
z = OPT['-z'] if '-z' in OPT else 'REG'
alg = OPT['-a'] if '-a' in OPT else 'smile'
os.environ[
"CUDA_VISIBLE_DEVICES"] = OPT['-g'] if '-g' in OPT else "0,1,2,3"
voc = utils.VocSmiles(init_from_file="data/chembl_voc.txt", max_len=100)
agent = GPT2Model(voc, n_layer=12)
agent.load_state_dict(
torch.load(params['pr_path'] + '.pkg', map_location=utils.dev))
prior = GPT2Model(voc, n_layer=12)
prior.load_state_dict(
torch.load(params['ft_path'] + '.pkg', map_location=utils.dev))
evolver = SmilesExplorer(agent, mutate=prior)
evolver.batch_size = BATCH_SIZE
evolver.epsilon = float(OPT.get('-e', '1e-2'))
evolver.sigma = float(OPT.get('-b', '0.00'))
evolver.scheme = OPT.get('-s', 'WS')
evolver.repeat = 1
keys = ['A2A', 'QED']
A2A = utils.Predictor('output/env/RF_%s_CHEMBL251.pkg' % z, type=z)
QED = utils.Property('QED')
# Chose the desirability function
objs = [A2A, QED]
if evolver.scheme == 'WS':
mod1 = utils.ClippedScore(lower_x=3, upper_x=10)
mod2 = utils.ClippedScore(lower_x=0, upper_x=1)
ths = [0.5, 0]
else:
mod1 = utils.ClippedScore(lower_x=3, upper_x=6.5)
mod2 = utils.ClippedScore(lower_x=0, upper_x=0.5)
ths = [0.99, 0]
mods = [mod1, mod2] if case == 'OBJ3' else [mod1, mod2]
evolver.env = utils.Env(objs=objs, mods=mods, keys=keys, ths=ths)
root = 'output/%s_%s' % (alg,
time.strftime('%y%m%d_%H%M%S', time.localtime()))
os.mkdir(root)
copy2(alg + '_ex.py', root)
copy2(alg + '.py', root)
# import evolve as agent
evolver.out = root + '/%s_%s_%s_%s_%.0e' % (alg, evolver.scheme, z, case,
evolver.epsilon)
evolver.fit(data_loader, test_loader=test_loader)
]:
# agent_path = 'output/%s_%s_256.pkg' % (path, method)
print(agent_path)
agent.load_state_dict(torch.load(agent_path))
z = 'REG'
keys = ['A2A']
A2A = utils.Predictor('output/env/RF_%s_CHEMBL251.pkg' % z, type=z)
QED = utils.Property('QED')
# Chose the desirability function
objs = [A2A, QED]
ths = [6.5, 0.0]
env = utils.Env(objs=objs, mods=None, keys=keys, ths=ths)
if method in ['atom']:
data = pd.read_table('data/ligand_mf_brics_test.txt')
# data = data.sample(BATCH_SIZE * 10)
data = torch.from_numpy(data.values).long().view(
len(data), voc.max_len, -1)
loader = DataLoader(data, batch_size=BATCH_SIZE)
out = '%s.txt' % agent_path
else:
data = pd.read_table('data/%s_test.txt' %
dataset).Input.drop_duplicates()
# data = data.sample(BATCH_SIZE * 10)
data = voc.encode([seq.split(' ')[:-1] for seq in data.values])
loader = DataLoader(data, batch_size=BATCH_SIZE)
TEAMPLATEFILE = 'template.html'
DATAFILE = 'data.json'
import utils
env = utils.Env()
def template2html(datapath=None,templatepath=None):
templatepath = templatepath if os.path.exists(datapath) else or os.path.join(env.home,DATAFILE)
if not os.path.exists(templatepath):
from string import Template
open (
s = Template('$who likes $what')
s.substitute(who='tim', what='kung pao')
'tim likes kung pao'
d = dict(who='tim')
Template('Give $who $100').substitute(d)
if __name__ == '__main__':
for z in ['REG']:
# Construct the environment with three predictors and desirability functions
keys = ['A1', 'A2A', 'ERG']
A1 = utils.Predictor('output/env/RF_%s_CHEMBL226.pkg' % z, type=z)
A2A = utils.Predictor('output/env/RF_%s_CHEMBL251.pkg' % z, type=z)
ERG = utils.Predictor('output/env/RF_%s_CHEMBL240.pkg' % z, type=z)
mod1 = utils.ClippedScore(lower_x=4, upper_x=6.5)
mod2 = utils.ClippedScore(lower_x=9, upper_x=6.5)
mod3 = utils.ClippedScore(lower_x=7.5, upper_x=5)
objs = [A1, A2A, ERG]
models = {
'output/lstm_ligand.pkg':
'benchmark/FINE-TUNE_%s_%s.tsv' % (z, case),
'output/lstm_chembl.pkg':
'benchmark/PRE-TRAIN_%s_%s.tsv' % (z, case)
}
for case in ['OBJ1', 'OBJ3']:
if case == 'OBJ3':
mods = [mod1, mod1, mod3]
else:
mods = [mod2, mod1, mod3]
env = utils.Env(objs=objs, mods=mods, keys=keys)
for input, output in models.items():
df = pd.DataFrame()
df['Smiles'] = sampling(input, output)
scores = env(df['Smiles'], is_smiles=True)
df.to_csv(output, index=False, sep='\t')
A2A = utils.Predictor('output/env/RF_%s_CHEMBL251.pkg' % z, type=z)
ERG = utils.Predictor('output/env/RF_%s_CHEMBL240.pkg' % z, type=z)
# Chose the desirability function
objs = [A1, A2A, ERG]
if scheme == 'WS':
mod1 = utils.ClippedScore(lower_x=3, upper_x=10)
mod2 = utils.ClippedScore(lower_x=10, upper_x=3)
ths = [0.5] * 3
else:
mod1 = utils.ClippedScore(lower_x=3, upper_x=6.5)
mod2 = utils.ClippedScore(lower_x=10, upper_x=6.5)
ths = [0.99] * 3
mods = [mod1, mod1, mod2] if case == 'OBJ3' else [mod2, mod1, mod2]
env = utils.Env(objs=objs, mods=mods, keys=keys, ths=ths)
root = 'output/%s_%s_%s_%s/'% (alg, case, scheme, time.strftime('%y%m%d_%H%M%S', time.localtime()))
os.mkdir(root)
copy2('models/rlearner.py', root)
copy2('trainer.py', root)
pr_path = 'output/lstm_chembl'
ft_path = 'output/lstm_ligand'
voc = utils.Voc(init_from_file="data/voc.txt")
agent = generator.Generator(voc)
agent.load_state_dict(torch.load(ft_path + '.pkg'))
prior = generator.Generator(voc)
prior.load_state_dict(torch.load(pr_path + '.pkg'))