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)
for agent_path in [
'benchmark/graph_PR_REG_OBJ1_0e+00.pkg',
'benchmark/graph_PR_REG_OBJ1_1e-01.pkg',
'benchmark/graph_PR_REG_OBJ1_1e-02.pkg',
'benchmark/graph_PR_REG_OBJ1_1e-03.pkg',
'benchmark/graph_PR_REG_OBJ1_1e-04.pkg',
'benchmark/graph_PR_REG_OBJ1_1e-05.pkg'
]:
# 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)
for i in tqdm(range(batch + 1)):
if i == 0:
if mod == 0: continue
tokens = netG.sample(batch)
else:
tokens = netG.sample(batch_size)
smiles = [voc.decode(s) for s in tokens]
samples.extend(smiles)
return samples
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':
def analysis_page():
# Methods
survey_file, config = request.args.get("survey_file"), request.args.get(
"config")
if (not request.method == "POST") and (not survey_file and not config):
return redirect(url_for("main"))
# Checking for files
do_analysis = False
if request.method == "GET":
if survey_file and config:
do_analysis = True
elif request.method == "POST" and (request.files["file"]
and request.files["config"]):
do_analysis = True
# Do analysis
if do_analysis:
# Saving files
if request.method == "POST":
save = save_file(survey_file=request.files["file"],
config_file=request.files["config"])
directory, filename, config_filename = (
save["Directory"],
save["File"],
save["Config"],
)
else:
directory, filename = os.path.split(survey_file)
config_filename = os.path.basename(config)
if filename.endswith(".xlsx"):
questions = list(
utils.parse_excel(os.path.join(directory, filename)).keys())
else:
questions = list(
utils.parse_csv(os.path.join(directory, filename)).keys())
types = utils.parse_config(os.path.join(directory, config_filename))
# Excel but incomplete config
if len(questions) != len(types):
session["TEMP_FOLDER"] = directory
predictor = utils.Predictor()
qn_dict = {}
for i, qn in enumerate(questions):
if i + 1 not in types.keys():
datatype = predictor.predict([qn])
qn_dict[i + 1] = (qn, datatype[0])
else:
qn_dict[i + 1] = (qn, types[i + 1])
questions_index = [(i[0], i[1][0], i[1][1])
for i in qn_dict.items()]
return render_template("config.html",
questions=questions_index,
error=None)
# Start analysis
try:
session["ANALYSIS"] = analyse.analyse(directory, filename,
config_filename)
except ValueError as e:
return render_template(
"error.html",
error=
"ValueError! Perhaps you chose a wrong category for your data",
error_no="500",
error_message=error_messages[500],
)
except Exception as e:
return render_template(
"error.html",
error=f"Unknown error: {str(e)}",
error_no="500",
error_message=error_messages[500],
)
graphs, clouds, numerical = [], [], []
for question, analysis in session["ANALYSIS"].items():
if analysis:
if analysis[0] == "categorical":
graphs.append([
question,
utils.pie(
question,
[x for x in analysis[1]["Percentages"].keys()],
[y for y in analysis[1]["Percentages"].values()],
),
])
elif analysis[0] == "openended":
clouds.append([question, analysis[1]])
elif analysis[0] == "numerical":
numerical.append([question, analysis[1]])
graphs = tuple(utils.chunk(graphs, 3))
clouds = tuple(utils.chunk(clouds, 2))
numerical = tuple(utils.chunk(numerical, 4))
return render_template(
"analysis.html",
graphs=graphs,
clouds=clouds,
numerical=numerical,
filename=filename,
path=os.path.split(directory)[1],
)
elif not request.files["file"]: # No excel
return render_template("upload.html", error="Missing Excel/CSV file!")
elif request.files[
"file"] and not request.files["config"]: # Excel but no config
save = save_file(survey_file=request.files["file"])
directory, filename = save["Directory"], save["File"]
session["TEMP_FOLDER"] = directory
if filename.endswith(".xlsx"):
questions = list(
utils.parse_excel(os.path.join(directory, filename)).keys())
else:
questions = list(
utils.parse_csv(os.path.join(directory, filename)).keys())
predictions = utils.Predictor().predict(questions)
questions_index = [(i + 1, question, predictions[i])
for i, question in enumerate(questions)]
return render_template("config.html",
questions=questions_index,
error=None)