def gen_items(space, covariate):
# Create an item bank to store the items.
# Setting the property "sizeHint" increases allocation efficiency
bank = gobject.new(oscats.ItemBank, sizeHint=N_ITEMS)
# Create the items
for i in range(N_ITEMS):
# First we create an IRT model container for our item
# We have to specify which dimensions to be used with the "dims" array
# (in this case, we use both of the dimensions of the space)
model = gobject.new(oscats.ModelL2p,
space=space,
dims=dims,
covariates=covariate)
# Then, set the parameters. Here there are 4:
# Discrimination on two dimensions, difficulty, and covariate coef.
model.set_param_by_name("Diff", oscats.oscats_rnd_normal(sqrt(3)))
model.set_param_by_name("Discr.Cont.1",
oscats.oscats_rnd_uniform_range(0, 1))
model.set_param_by_name("Discr.Cont.2",
oscats.oscats_rnd_uniform_range(0, 2))
model.set_param_by_name(COVARIATE_NAME,
oscats.oscats_rnd_uniform_range(0.5, 1.5))
# Create an item based on this model
item = gobject.new(oscats.Item)
item.set_model(item.get_default_model(), model)
# Add the item to the item bank
bank.add_item(item)
# Since Python is garbage collected, we don't have to worry about
# reference counting.
return bank
def gen_items(space, covariate) :
# Create an item bank to store the items.
# Setting the property "sizeHint" increases allocation efficiency
bank = gobject.new(oscats.ItemBank, sizeHint=N_ITEMS)
# Create the items
for i in range(N_ITEMS) :
# First we create an IRT model container for our item
# We have to specify which dimensions to be used with the "dims" array
# (in this case, we use both of the dimensions of the space)
model = gobject.new(oscats.ModelL2p, space=space, dims=dims, covariates=covariate)
# Then, set the parameters. Here there are 4:
# Discrimination on two dimensions, difficulty, and covariate coef.
model.set_param_by_name("Diff", oscats.oscats_rnd_normal(sqrt(3)))
model.set_param_by_name("Discr.Cont.1",
oscats.oscats_rnd_uniform_range(0, 1))
model.set_param_by_name("Discr.Cont.2",
oscats.oscats_rnd_uniform_range(0, 2))
model.set_param_by_name(COVARIATE_NAME,
oscats.oscats_rnd_uniform_range(0.5, 1.5))
# Create an item based on this model
item = gobject.new(oscats.Item)
item.set_model(item.get_default_model(), model)
# Add the item to the item bank
bank.add_item(item)
# Since Python is garbage collected, we don't have to worry about
# reference counting.
return bank
def gen_examinees(space):
dim = space.get_dim_by_name("Cont.1")
ret = []
for i in range(N_EXAMINEES):
theta = oscats.Point(space=space)
theta.set_cont(dim, oscats.oscats_rnd_normal(1))
e = oscats.Examinee()
e.set_sim_theta(theta)
ret.append(e)
return ret
def gen_examinees(space):
dim = space.get_dim_by_name("Cont.1")
ret = []
for i in range(N_EXAMINEES):
theta = oscats.Point(space=space)
theta.set_cont(dim, oscats.oscats_rnd_normal(1))
e = oscats.Examinee()
e.set_sim_theta(theta)
ret.append(e)
return ret
def gen_examinees(space) :
dim = space.get_dim_by_name("Cont.1")
ret = []
for i in range(N_EXAMINEES) :
# Latent IRT ability parameter. This is a one-dimensional test.
theta = oscats.Point(space=space)
# Sample the ability from N(0,1) distribution
theta.set_cont(dim, oscats.oscats_rnd_normal(1))
# Create a new examinee
e = gobject.new(oscats.Examinee)
# Set the examinee's true (simulated) ability
e.set_sim_theta(theta)
ret.append(e)
return ret
def gen_examinees(space):
dim = space.get_dim_by_name("Cont.1")
ret = []
for i in range(N_EXAMINEES):
# Latent IRT ability parameter. This is a one-dimensional test.
theta = oscats.Point(space=space)
# Sample the ability from N(0,1) distribution
theta.set_cont(dim, oscats.oscats_rnd_normal(1))
# Create a new examinee
e = gobject.new(oscats.Examinee)
# Set the examinee's true (simulated) ability
e.set_sim_theta(theta)
ret.append(e)
return ret
def gen_items(space) :
# Create an item bank to store the items.
# Setting the property "sizeHint" increases allocation efficiency
bank = gobject.new(oscats.ItemBank, sizeHint=N_ITEMS)
for i in range(N_ITEMS) :
# First we create an IRT model container for our item
# Defaults to unidimensional, using the first dimension of space
model = gobject.new(oscats.ModelL1p, space=space)
# Then, set the parameters. Here there is only one, the difficulty (b).
model.set_param_by_index(0, oscats.oscats_rnd_normal(1))
# Create an item based on this model
item = gobject.new(oscats.Item)
item.set_model(item.get_default_model(), model)
# Add the item to the item bank
bank.add_item(item)
# Since Python is garbage collected, we don't have to worry about
# reference counting.
return bank
def gen_items(space):
# Create an item bank to store the items.
# Setting the property "sizeHint" increases allocation efficiency
bank = gobject.new(oscats.ItemBank, sizeHint=N_ITEMS)
for i in range(N_ITEMS):
# First we create an IRT model container for our item
# Defaults to unidimensional, using the first dimension of space
model = gobject.new(oscats.ModelL1p, space=space)
# Then, set the parameters. Here there is only one, the difficulty (b).
model.set_param_by_index(0, oscats.oscats_rnd_normal(1))
# Create an item based on this model
item = gobject.new(oscats.Item)
item.set_model(item.get_default_model(), model)
# Add the item to the item bank
bank.add_item(item)
# Since Python is garbage collected, we don't have to worry about
# reference counting.
return bank
"\t".join([ "%s.theta\t%s.alpha" for name in test_names]) + "\n")
for e in examinees :
tmp = e.get_theta_by_name("trueTheta")
dim = contSpace.get_dim_by_name("Cont.1")
out.write("%s\t%g\t" % (e.get_property("id"), tmp.get_cont(dim)))
tmp = e.get_theta_by_name("trueAlpha")
dim = binSpace.get_dim_by_name("Bin.1")
for k in range(N_ATTR) :
if tmp.get_bin(dim) : out.write("1")
else : out.write("0")
dim += 1
# Choose initial theta and alpha for this examinee
# We'll use the same initial abilities for all four tests
dim = contSpace.get_dim_by_name("Cont.1")
initTheta.set_cont(dim, oscats.oscats_rnd_normal(sqrt(0.05))-0.37)
dim = binSpace.get_dim_by_name("Bin.1")
for k in range(N_ATTR) :
initAlpha.set_bin(dim, oscats.oscats_rnd_uniform() < 0.5) ; dim += 1
# Initialize ability estimates for this examinee
e.set_theta_by_name("estTheta", oscats.Point(space=contSpace))
e.set_theta_by_name("estAlpha", oscats.Point(space=binSpace))
for test in tests :
# Reset initial latent ability for this test
e.get_theta_by_name("estTheta").copy(initTheta)
e.get_theta_by_name("estAlpha").copy(initAlpha)
# Do the administration!
test.administer(e)
# Here, we'll just use one object over again for each administration.
e = oscats.Examinee(covariates=covariate)
sim_theta = oscats.Point(space=space)
est_theta = oscats.Point(space=space)
e.set_sim_theta(sim_theta)
e.set_est_theta(est_theta)
print "Administering."
for I in range(len(grid)) :
sim_theta.set_cont(dims[0], grid[I])
for J in range(len(grid)) :
stdout.write("\rAt grid point (%g, %g) ...\n" % (grid[I], grid[J]))
sim_theta.set_cont(dims[1], grid[J])
for i in range(N_EXAMINEES) :
# "covariate" is the same OscatsCovariates object used by "e"
covariate.set_by_name(COVARIATE_NAME, oscats.oscats_rnd_normal(1))
stdout.write("\r%d" % i) ; stdout.flush()
for j in range(len(tests)) :
# Reset initial theta: Random start ~ N_2(0, I_2)
est_theta.set_cont(dims[0], oscats.oscats_rnd_normal(1))
est_theta.set_cont(dims[1], oscats.oscats_rnd_normal(1))
# Administer the test
tests[j].administer(e)
# Calculate error (Euclidean distance between sim/est theta)
err_sums[j][I*len(grid)+J] += distance(sim_theta, est_theta);
# Report results
out = open("ex02-results.txt", "w")
out.write("D1\tD2\t" + "\t".join(test_names) + "\n")
for I in range(len(grid)) :
for J in range(len(grid)) :
# Here, we'll just use one object over again for each administration.
e = oscats.Examinee(covariates=covariate)
sim_theta = oscats.Point(space=space)
est_theta = oscats.Point(space=space)
e.set_sim_theta(sim_theta)
e.set_est_theta(est_theta)
print "Administering."
for I in range(len(grid)):
sim_theta.set_cont(dims[0], grid[I])
for J in range(len(grid)):
stdout.write("\rAt grid point (%g, %g) ...\n" % (grid[I], grid[J]))
sim_theta.set_cont(dims[1], grid[J])
for i in range(N_EXAMINEES):
# "covariate" is the same OscatsCovariates object used by "e"
covariate.set_by_name(COVARIATE_NAME, oscats.oscats_rnd_normal(1))
stdout.write("\r%d" % i)
stdout.flush()
for j in range(len(tests)):
# Reset initial theta: Random start ~ N_2(0, I_2)
est_theta.set_cont(dims[0], oscats.oscats_rnd_normal(1))
est_theta.set_cont(dims[1], oscats.oscats_rnd_normal(1))
# Administer the test
tests[j].administer(e)
# Calculate error (Euclidean distance between sim/est theta)
err_sums[j][I * len(grid) + J] += distance(
sim_theta, est_theta)
# Report results
out = open("ex02-results.txt", "w")
out.write("D1\tD2\t" + "\t".join(test_names) + "\n")
"\t".join(["%s.theta\t%s.alpha" for name in test_names]) + "\n")
for e in examinees:
tmp = e.get_theta_by_name("trueTheta")
dim = contSpace.get_dim_by_name("Cont.1")
out.write("%s\t%g\t" % (e.get_property("id"), tmp.get_cont(dim)))
tmp = e.get_theta_by_name("trueAlpha")
dim = binSpace.get_dim_by_name("Bin.1")
for k in range(N_ATTR):
if tmp.get_bin(dim): out.write("1")
else: out.write("0")
dim += 1
# Choose initial theta and alpha for this examinee
# We'll use the same initial abilities for all four tests
dim = contSpace.get_dim_by_name("Cont.1")
initTheta.set_cont(dim, oscats.oscats_rnd_normal(sqrt(0.05)) - 0.37)
dim = binSpace.get_dim_by_name("Bin.1")
for k in range(N_ATTR):
initAlpha.set_bin(dim,
oscats.oscats_rnd_uniform() < 0.5)
dim += 1
# Initialize ability estimates for this examinee
e.set_theta_by_name("estTheta", oscats.Point(space=contSpace))
e.set_theta_by_name("estAlpha", oscats.Point(space=binSpace))
for test in tests:
# Reset initial latent ability for this test
e.get_theta_by_name("estTheta").copy(initTheta)
e.get_theta_by_name("estAlpha").copy(initAlpha)
