def test_no_token_get_data(self):
with pytest.raises(Exception):
dh.get_data({}, "short_term")
assert len(dh.get_top_track_names()) == 0
with pytest.raises(Exception):
dh.get_data({}, "medium_term")
assert len(dh.get_top_artist_names()) == 0
with pytest.raises(Exception):
dh.get_data({}, "long_term")
assert len(dh.get_top_track_names()) == 0
def init(config, run):
# general init
args = SimpleNamespace(**config)
args = assertions.validate_args(args)
mlh.seed_all(args.seed)
args._run = run
args.wandb = wandb
# init scheduler
args.partition_scheduler = schedules.get_partition_scheduler(args)
args.partition = util.get_partition(args)
# init data
train_data_loader, test_data_loader = get_data(args)
args.train_data_loader = train_data_loader
args.test_data_loader = test_data_loader
# init model
model = get_model(train_data_loader, args)
# init optimizer
model.init_optimizer()
return model, args
def train(args):
# read data
train_data_loader, test_data_loader = get_data(args)
# attach data to args
args.train_data_loader = train_data_loader
args.test_data_loader = test_data_loader
# Make models
model = get_model(train_data_loader, args)
# Make optimizer
if args.loss in DUAL_OBJECTIVES:
optimizer_phi = torch.optim.Adam(
(params for name, params in model.named_parameters()
if args.phi_tag in name),
lr=args.lr)
optimizer_theta = torch.optim.Adam(
(params for name, params in model.named_parameters()
if args.theta_tag in name),
lr=args.lr)
else:
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
#for epoch in range(args.epochs):
for epoch in tqdm(range(args.epochs)):
if mlh.is_schedule_update_time(epoch, args):
args.partition = args.partition_scheduler(model, args)
if len(args.Y_ori) % args.increment_update_frequency == 0 and len(
args.Y_ori) > 1:
args.schedule_update_frequency = args.schedule_update_frequency + 1
print("args.schedule_update_frequency=",
args.schedule_update_frequency)
if args.loss in DUAL_OBJECTIVES:
train_logpx, train_elbo, train_tvo_log_evidence = model.train_epoch_dual_objectives(
train_data_loader, optimizer_phi, optimizer_theta, epoch=epoch)
else:
# addl recording within model.base
train_logpx, train_elbo, train_tvo_log_evidence = model.train_epoch_single_objective(
train_data_loader, optimizer, epoch=epoch)
log_scalar(train_elbo=train_elbo,
train_logpx=train_logpx,
train_tvo_log_evidence=train_tvo_log_evidence,
step=epoch)
# store the information
args.betas_all = np.vstack((args.betas_all,
np.reshape(format_input(args.partition),
(1, args.K + 1))))
args.logtvopx_all = np.append(args.logtvopx_all,
train_tvo_log_evidence)
if mlh.is_gradient_time(epoch, args):
# Save grads
grad_variance = util.calculate_grad_variance(model, args)
log_scalar(grad_variance=grad_variance, step=epoch)
if mlh.is_test_time(epoch, args):
test_logpx, test_kl = model.evaluate_model_and_inference_network(
test_data_loader, epoch=epoch)
log_scalar(test_logpx=test_logpx, test_kl=test_kl, step=epoch)
if mlh.is_checkpoint_time(epoch, args):
opt = [optimizer_phi, optimizer_theta
] if args.loss in DUAL_OBJECTIVES else [optimizer]
save_checkpoint(model, epoch, train_elbo, train_logpx, opt, args)
# ------ end of training loop ---------
opt = [optimizer_phi, optimizer_theta
] if args.loss in DUAL_OBJECTIVES else [optimizer]
save_checkpoint(model, args.epochs, train_elbo, train_logpx, opt, args)
if args.train_only:
test_logpx, test_kl = 0, 0
results = {
"test_logpx": test_logpx,
"test_kl": test_kl,
"train_logpx": train_logpx,
"train_elbo": train_elbo,
"train_tvo_px": train_tvo_log_evidence,
"average_y":
args.average_y, # average tvo_logpx within this bandit iteration
"X": args.X_ori, # this is betas
# this is utility score y=f(betas)= ave_y[-1] - ave_y[-2]
"Y": args.Y_ori
}
return results, model
def test_write_data(init, test_input):
init.add_user(test_input)
wrote_list = write_data_out(init, URL)
assert wrote_list.gib_list() == get_data(URL).gib_list()
def test_get_data_fail(init):
with pytest.raises(Exception):
get_data("emptydb.json")
def test_get_data(init):
assert init.gib_list() == get_data(URL).gib_list()
def train(args):
# read data
train_data_loader, test_data_loader = get_data(args)
# attach data to args
args.train_data_loader = train_data_loader
args.test_data_loader = test_data_loader
# Make models
model = get_model(train_data_loader, args)
args.train_data_loader = train_data_loader
args.test_data_loader = test_data_loader
# Make optimizer
if args.loss in DUAL_LIST:
optimizer_phi = torch.optim.Adam(
(params for name, params in model.named_parameters()
if 'encoder' in name),
lr=args.lr)
optimizer_theta = torch.optim.Adam(
(params for name, params in model.named_parameters()
if 'decoder' in name),
lr=args.lr)
else:
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
for epoch in range(args.epochs):
_record = mlh.is_record_time(epoch, args)
if args.loss in DUAL_LIST:
train_logpx, train_elbo = model.train_epoch_dual_objectives(
train_data_loader,
optimizer_phi,
optimizer_theta,
record=_record)
else:
train_logpx, train_elbo = model.train_epoch_single_objective(
train_data_loader, optimizer, record=_record)
log_scalar(train_elbo=train_elbo, train_logpx=train_logpx, step=epoch)
if mlh.is_test_time(epoch, args):
args.test_time = True
test_logpx, test_kl = model.evaluate_model_and_inference_network(
test_data_loader)
log_scalar(test_logpx=test_logpx, test_kl=test_kl, step=epoch)
args.test_time = False
if mlh.is_checkpoint_time(epoch, args):
opt = [optimizer_phi, optimizer_theta
] if args.loss in DUAL_LIST else [optimizer]
save_checkpoint(model, epoch, train_elbo, train_logpx, opt, args)
if mlh.is_schedule_update_time(epoch, args):
args.partition = args.partition_scheduler(model, args)
# ------ end of training loop ---------
if args.train_only:
test_logpx, test_kl = 0, 0
results = {
"test_logpx": test_logpx,
"test_kl": test_kl,
"train_logpx": train_logpx,
"train_elbo": train_elbo
}
return results, model
from typing import Dict
from fastapi import FastAPI
from src.data_handler import get_data, write_data_out
users_list = get_data() # initializing list from db
app = FastAPI()
@app.get("/users")
def get_users(usr_list: Dict, db=users_list):
usrs = usr_list["users"]
if len(usrs) == 1:
return db.gib_list(usrs[0])
found = []
for i in range(len(usrs)):
found.append(db.gib_list(usrs[i]))
return found
@app.post("/user")
def post_user(user: Dict, db=users_list):
added_user = db.add_user(user["user"])
write_data_out(db)
return db.gib_list(added_user.name)
@app.post("/iou")
def post_iou(info: Dict, db=users_list):
lender = info["lender"]
borrower = info["borrower"]
amount = info["amount"]
