#print(real_diff)
#print(real_slope)
#print(responses)
num_models = len(set(models))
num_items = len(set(items))
#print(num_items, num_models)
models = torch.tensor(models, dtype=torch.long, device=device)
items = torch.tensor(items, dtype=torch.long, device=device)
responses = torch.tensor(responses, dtype=torch.float, device=device)
# 3. define model and guide accordingly
m1v = OneParamLog("vague", device, num_items, num_models, args.verbose)
m1h = OneParamLog("hierarchical", device, num_items, num_models, args.verbose)
m2v = TwoParamLog("vague", device, num_items, num_models, args.verbose)
m2h = TwoParamLog("hierarchical", device, num_items, num_models, args.verbose)
pyro.enable_validation(True)
for m in [m1v, m2v, m1h, m2h]:
# 4. fit irt model with svi, trace-elbo loss
m.fit(models, items, responses, args.num_epochs)
# 5. once model is fit, write outputs (diffs and thetas) to disk,
# retaining original modelIDs and itemIDs so we can use them
for name in pyro.get_param_store().get_all_param_names():
if name not in ['loc_diff', 'loc_ability', 'loc_slope']:
continue
print(name)
val = pyro.param(name).data.numpy()
def test_num_subjects(self):
with self.assertRaises(ValueError):
m = TwoParamLog("vague", "cpu", 100, -100, False)
def test_device(self):
with self.assertRaises(ValueError):
m = TwoParamLog("vague", "zpu", 100, 100, False)
def test_priors(self):
with self.assertRaises(ValueError):
m = TwoParamLog("testing", "cpu", 100, 100, False)
iID += 1
uID += 1
num_models = len(set(models))
num_items = len(set(items))
print(num_items, num_models)
models = torch.tensor(models, dtype=torch.long, device=device)
items = torch.tensor(items, dtype=torch.long, device=device)
responses = torch.tensor(responses, dtype=torch.float, device=device)
# 3. define model and guide accordingly
if args.model == '1PL':
m = OneParamLog(args.priors, device, num_items, num_models, verbose=args.verbose)
elif args.model == '2PL':
m = TwoParamLog(args.priors, device, num_items, num_models, verbose=args.verbose)
# 4. fit irt model with svi, trace-elbo loss
m.fit(models, items, responses, args.num_epochs)
# 5. once model is fit, write outputs (diffs and thetas) to disk,
# retaining original modelIDs and itemIDs so we can use them
for name in pyro.get_param_store().get_all_param_names():
print(name)
if args.gpu:
val = pyro.param(name).data.cpu().numpy()
else:
val = pyro.param(name).data.numpy()
if args.verbose:
models = torch.tensor(models, dtype=torch.long, device=device)
items = torch.tensor(items, dtype=torch.long, device=device)
responses = torch.tensor(responses, dtype=torch.float, device=device)
# 3. define model and guide accordingly
if args.model == "1PL":
m = OneParamLog(args.priors,
device,
num_items,
num_subjects,
verbose=args.verbose)
elif args.model == "2PL":
m = TwoParamLog(args.priors,
device,
num_items,
num_subjects,
verbose=args.verbose)
# 4. fit irt model with svi, trace-elbo loss
m.fit(models, items, responses, args.num_epochs)
# 5. once model is fit, write outputs (diffs and thetas) to disk,
# retaining original modelIDs and itemIDs so we can use them
for name in pyro.get_param_store().get_all_param_names():
print(name)
if args.gpu:
val = pyro.param(name).data.cpu().numpy()
else:
val = pyro.param(name).data.numpy()