def predict():
request_body = request.get_json()
instance = request_body['instance']
output = inference.estimate(instance)
return jsonify(output)
def direct_process(args, net, device: str = 'cpu'):
# Init.
coeffdict = read_coeff(args.coeff)
beta = [a for a in args.alpha]
# Instantiate dataloader
stereo_dataset = InferenceRun(root=args.root, pair=False, use_stereo=True)
stereo_dataloader = DataLoader(stereo_dataset,
batch_size=1,
shuffle=False,
num_workers=8,
pin_memory=True)
# Processing dataloader
for images, floname in tqdm(stereo_dataloader,
ncols=100,
leave=True,
unit='pair',
desc=f'Evaluating {args.root}'):
images = [image.to(device) for image in images] # Add to device
flow_cal = [
_stereo_cal(estimate(net, images[i], images[i + 1],
tensor=False), coeffdict[naming.capitalize()],
args.window_size, 1 / args.fps, True)
for i, naming in enumerate(['left', 'right'])
]
stereo_flow = willert(flow_cal, args.theta, beta)
flosave = str(floname[0]) + '_2d3c.flo'
write_flow(stereo_flow, os.path.join(args.save, "stereo", flosave))
MODEL_NAME = 'synth_custom_regressor'
MODEL_VERSION = 'v1'
def compute_rmse(estimates, actual):
rmse = math.sqrt(((pd.Series(estimates) - actual)**2).sum() / len(actual))
return rmse
test_data = pd.read_csv("data/test-data.csv",
header=None,
names="key,x,y,alpha,beta,target".split(','))
test_instances = list(
test_data.apply(lambda row: {
'x': row['x'],
'y': row['y'],
'alpha': row['alpha'],
'beta': row['beta']
},
axis=1))
estimates = inference.estimate(instances=test_instances,
project=PROJECT,
model_name=MODEL_NAME,
version=MODEL_VERSION)
rmse = compute_rmse(estimates, test_data.target)
print("Test RMSE: {}".format(round(rmse, 3)))
import inference
instance = {
'is_male': 'True',
'mother_age': 26.0,
'mother_race': 'Asian Indian',
'plurality': 1.0,
'gestation_weeks': 39,
'mother_married': 'True',
'cigarette_use': 'False',
'alcohol_use': 'False'
}
output = inference.estimate(instance)
print(output)