def do_classification(feature_data, predict, params):
length = params[0]['max_length']
x, m = batch.make_batch(feature_data, length, length / 2)
#decision = predict(np.expand_dims(feature_data,axis=0).astype('float32'), np.ones(shape=(1,feature_data.shape[0])))
decision = predict(x, m)
pred_label = np.argmax(np.sum(decision, axis=0), axis=-1)
return batch.labels[pred_label]
def train(args, data, model, criterion, optimizer):
[
prot_fea_list, drug_node_list, drug_edge_list, drug_n2n_list,
drug_e2n_list, label_list
] = data
prot_data = [prot_fea_list]
drug_data = [drug_node_list, drug_edge_list, drug_n2n_list, drug_e2n_list]
label_data = [label_list]
n_data = len(label_list)
n_step = int(n_data / args.n_batch) + 1
batch_idx_list = split(list(range(n_data)), n_step, shuffle=True)
total_loss = 0
model.train()
for i, batch_idx in enumerate(batch_idx_list):
batch_data = batch.make_batch(args,
data=[prot_data, drug_data, label_data],
idx=batch_idx)
batch_prot_data, batch_drug_data, batch_label_data = batch_data
optimizer.zero_grad()
pred = model(batch_prot_data, batch_drug_data)
loss = criterion(pred.squeeze(), batch_label_data)
loss.backward()
optimizer.step()
total_loss += loss.data
return total_loss / n_step
def do_classification(feature_data, predict, params):
length = params[0]['max_length']
x, m = batch.make_batch(feature_data,length,length/2)
#decision = predict(np.expand_dims(feature_data,axis=0).astype('float32'), np.ones(shape=(1,feature_data.shape[0])))
decision = predict(x, m)
pred_label = np.argmax(np.sum(decision,axis=0), axis = -1)
return batch.labels[pred_label]
def do_classification(feature_data, predict, params):
length = params[0]['max_length']
x, m = batch.make_batch(feature_data, length, length / 2)
x = batch.make_context(x, 15)
#decision = predict(np.expand_dims(feature_data,axis=0).astype('float32'), np.ones(shape=(1,feature_data.shape[0])))
decision = predict(x)
return decision
def do_classification(feature_data, predict, params):
length = params[0]['max_length']
x, m = batch.make_batch(feature_data,length,length/2)
x=batch.make_context(x,15)
#decision = predict(np.expand_dims(feature_data,axis=0).astype('float32'), np.ones(shape=(1,feature_data.shape[0])))
decision = predict(x)
return decision
def do_classification(feature_data, predict, params):
'''
input feature_data
return classification results
'''
x, _ = batch.make_batch(feature_data, 15, 5)
decision = predict(x.reshape((x.shape[0], -1)))
return decision
def do_classification(feature_data, predict, params):
'''
input feature_data
return classification results
'''
x, _ = batch.make_batch(feature_data,15,5)
decision = predict(x.reshape((x.shape[0],-1)))
return decision
def test(args, data, model, criterion, val):
if val:
shuffle = True
n_mc_step = 1
else:
shuffle = False
n_mc_step = args.n_mc_step
[
prot_fea_list, drug_node_list, drug_edge_list, drug_n2n_list,
drug_e2n_list, label_list
] = data
prot_data = [prot_fea_list]
drug_data = [drug_node_list, drug_edge_list, drug_n2n_list, drug_e2n_list]
label_data = [label_list]
n_data = len(label_list)
n_step = int(n_data / args.n_batch) + 1
batch_idx_list = split(list(range(n_data)), n_step, shuffle=shuffle)
y_pred_list = []
label_list = []
score_list = []
total_loss = 0
model.train()
for mc_idx in range(n_mc_step):
buf_y_pred_list = []
buf_label_list = []
for i, batch_idx in enumerate(batch_idx_list):
batch_data = batch.make_batch(
args, data=[prot_data, drug_data, label_data], idx=batch_idx)
batch_prot_data, batch_drug_data, batch_label_data = batch_data
y_pred = model(batch_prot_data, batch_drug_data)
loss = criterion(y_pred.squeeze(), batch_label_data)
total_loss += loss.data
for yp in y_pred.squeeze().cpu().detach().numpy():
buf_y_pred_list.append(yp)
for l in batch_label_data.squeeze().cpu().detach().numpy():
buf_label_list.append(l)
y_pred_list.append(buf_y_pred_list)
label_list.append(buf_label_list)
score_list.append(cal_roc_auc(buf_label_list, buf_y_pred_list))
y_pred_list = np.array(y_pred_list)
label_list = np.array(label_list)
#print("P = {}".format(model.prot_encoder.prot_fc1.ConDrop.p))
return (total_loss / n_step / n_mc_step, np.mean(score_list), label_list,
y_pred_list)
def do_classification(feature_data, predict, params):
'''
input feature_data
return classification results
'''
# ???
#import pdb; pdb.set_trace()
x, _ = batch.make_batch(feature_data,params[0]['max_length'],params[0]['max_length'])
x = reshape(x)
decision = predict(x)
return decision
def test(args, data, model, criterion, val):
if val:
shuffle = True
else:
shuffle = False
[
prot_fea_list, drug_node_list, drug_edge_list, drug_n2n_list,
drug_e2n_list, label_list
] = data
prot_data = [prot_fea_list]
drug_data = [drug_node_list, drug_edge_list, drug_n2n_list, drug_e2n_list]
label_data = [label_list]
n_data = len(label_list)
n_step = int(n_data / args.n_batch) + 1
batch_idx_list = split(list(range(n_data)), n_step, shuffle=shuffle)
y_pred_list = []
label_list = []
total_loss = 0
model.eval()
for i, batch_idx in enumerate(batch_idx_list):
batch_data = batch.make_batch(args,
data=[prot_data, drug_data, label_data],
idx=batch_idx)
batch_prot_data, batch_drug_data, batch_label_data = batch_data
y_pred = model(batch_prot_data, batch_drug_data)
loss = criterion(y_pred.squeeze(), batch_label_data)
total_loss += loss.data
for yp in y_pred.squeeze().cpu().detach().numpy():
y_pred_list.append(yp)
for l in batch_label_data.squeeze().cpu().detach().numpy():
label_list.append(l)
score = cal_roc_auc(label_list, y_pred_list)
y_pred_list = np.array(y_pred_list)
label_list = np.array(label_list)
return (total_loss / n_step, score, label_list, y_pred_list)
output, main = solve_system(system)
#postprocess
if output is not None:
main = postprocess_data(system, output)
if main == 1: break
else: break
if option == 2:
#run a batch file
systems, type, main = make_batch(system)
if main != 1:
outputs, main = solve_batch(systems, type)
if outputs is not None:
main = postprocess_batch(type, systems, outputs)
if main == 1: break
if option == 3:
#Load an existing output file
while (1):