def _fit(self, dataset):
self.dataset = dataset
self.masker = self.masker or dataset.masker
if not isinstance(self.masker, NiftiMasker):
raise ValueError(
f"A {type(self.masker)} mask has been detected. "
"Only NiftiMaskers are allowed for this Estimator. "
"This is because aggregation, such as averaging values across ROIs, "
"will produce invalid results.")
est = pymare.estimators.StoufferCombinationTest()
if self.use_sample_size:
sample_sizes = np.array(
[np.mean(n) for n in self.inputs_["sample_sizes"]])
weights = np.sqrt(sample_sizes)
weight_maps = np.tile(weights,
(self.inputs_["z_maps"].shape[1], 1)).T
pymare_dset = pymare.Dataset(y=self.inputs_["z_maps"],
v=weight_maps)
else:
pymare_dset = pymare.Dataset(y=self.inputs_["z_maps"])
est.fit_dataset(pymare_dset)
est_summary = est.summary()
results = {
"z":
_boolean_unmask(est_summary.z.squeeze(),
self.inputs_["aggressive_mask"]),
"p":
_boolean_unmask(est_summary.p.squeeze(),
self.inputs_["aggressive_mask"]),
}
return results
def _fit(self, dataset):
self.dataset = dataset
self.masker = self.masker or dataset.masker
if not isinstance(self.masker, NiftiMasker):
LGR.warning(
f"A {type(self.masker)} mask has been detected. "
"Masks which average across voxels will likely produce biased results when used "
"with this Estimator.")
est = pymare.estimators.VarianceBasedLikelihoodEstimator(
method=self.method)
pymare_dset = pymare.Dataset(y=self.inputs_["beta_maps"],
v=self.inputs_["varcope_maps"])
est.fit_dataset(pymare_dset)
est_summary = est.summary()
results = {
"tau2":
_boolean_unmask(est_summary.tau2.squeeze(),
self.inputs_["aggressive_mask"]),
"z":
_boolean_unmask(est_summary.get_fe_stats()["z"].squeeze(),
self.inputs_["aggressive_mask"]),
"p":
_boolean_unmask(est_summary.get_fe_stats()["p"].squeeze(),
self.inputs_["aggressive_mask"]),
"est":
_boolean_unmask(est_summary.get_fe_stats()["est"].squeeze(),
self.inputs_["aggressive_mask"]),
}
return results
def _fit(self, dataset):
self.dataset = dataset
self.masker = self.masker or dataset.masker
sample_sizes = np.array(
[np.mean(n) for n in self.inputs_["sample_sizes"]])
n_maps = np.tile(sample_sizes,
(self.inputs_["beta_maps"].shape[1], 1)).T
pymare_dset = pymare.Dataset(y=self.inputs_["beta_maps"], n=n_maps)
est = pymare.estimators.SampleSizeBasedLikelihoodEstimator(
method=self.method)
est.fit_dataset(pymare_dset)
est_summary = est.summary()
results = {
"tau2":
_boolean_unmask(est_summary.tau2.squeeze(),
self.inputs_["aggressive_mask"]),
"z":
_boolean_unmask(est_summary.get_fe_stats()["z"].squeeze(),
self.inputs_["aggressive_mask"]),
"p":
_boolean_unmask(est_summary.get_fe_stats()["p"].squeeze(),
self.inputs_["aggressive_mask"]),
"est":
_boolean_unmask(est_summary.get_fe_stats()["est"].squeeze(),
self.inputs_["aggressive_mask"]),
}
return results
def _fit(self, dataset):
self.dataset = dataset
self.masker = self.masker or dataset.masker
if not isinstance(self.masker, NiftiMasker):
LGR.warning(
f"A {type(self.masker)} mask has been detected. "
"Masks which average across voxels will likely produce biased results when used "
"with this Estimator.")
pymare_dset = pymare.Dataset(y=self.inputs_["beta_maps"],
v=self.inputs_["varcope_maps"])
est = pymare.estimators.WeightedLeastSquares(tau2=self.tau2)
est.fit_dataset(pymare_dset)
est_summary = est.summary()
# tau2 is an float, not a map, so it can't go in the results dictionary
results = {
"z":
_boolean_unmask(est_summary.get_fe_stats()["z"].squeeze(),
self.inputs_["aggressive_mask"]),
"p":
_boolean_unmask(est_summary.get_fe_stats()["p"].squeeze(),
self.inputs_["aggressive_mask"]),
"est":
_boolean_unmask(est_summary.get_fe_stats()["est"].squeeze(),
self.inputs_["aggressive_mask"]),
}
return results
def _fit(self, dataset):
if self.use_sample_size:
sample_sizes = np.array([np.mean(n) for n in self.inputs_["sample_sizes"]])
weights = np.sqrt(sample_sizes)
weight_maps = np.tile(weights, (self.inputs_["z_maps"].shape[1], 1)).T
pymare_dset = pymare.Dataset(y=self.inputs_["z_maps"], v=weight_maps)
else:
pymare_dset = pymare.Dataset(y=self.inputs_["z_maps"])
est = pymare.estimators.Stouffers(input="z")
est.fit(pymare_dset)
est_summary = est.summary()
results = {
"z": est_summary.z,
"p": est_summary.p,
}
return results
def _fit(self, dataset):
pymare_dset = pymare.Dataset(y=self.inputs_["z_maps"])
est = pymare.estimators.FisherCombinationTest()
est.fit_dataset(pymare_dset)
est_summary = est.summary()
results = {
"z": est_summary.z,
"p": est_summary.p,
}
return results
def _fit(self, dataset):
est = pymare.estimators.Hedges()
pymare_dset = pymare.Dataset(y=self.inputs_["beta_maps"], v=self.inputs_["varcope_maps"])
est.fit(pymare_dset)
est_summary = est.summary()
results = {
"tau2": est_summary.tau2,
"z": est_summary.get_fe_stats()["z"],
"p": est_summary.get_fe_stats()["p"],
"est": est_summary.get_fe_stats()["est"],
}
return results
def _fit(self, dataset):
est = pymare.estimators.VarianceBasedLikelihoodEstimator(method=self.method)
pymare_dset = pymare.Dataset(y=self.inputs_["beta_maps"], v=self.inputs_["varcope_maps"])
est.fit(pymare_dset)
est_summary = est.summary()
results = {
"tau2": est_summary.tau2,
"z": est_summary.get_fe_stats()["z"],
"p": est_summary.get_fe_stats()["p"],
"est": est_summary.get_fe_stats()["est"],
}
return results
def _fit(self, dataset):
sample_sizes = np.array([np.mean(n) for n in self.inputs_["sample_sizes"]])
n_maps = np.tile(sample_sizes, (self.inputs_["beta_maps"].shape[1], 1)).T
pymare_dset = pymare.Dataset(y=self.inputs_["beta_maps"], n=n_maps)
est = pymare.estimators.SampleSizeBasedLikelihoodEstimator(method=self.method)
est.fit(pymare_dset)
est_summary = est.summary()
results = {
"tau2": est_summary.tau2,
"z": est_summary.get_fe_stats()["z"],
"p": est_summary.get_fe_stats()["p"],
"est": est_summary.get_fe_stats()["est"],
}
return results
def _fit(self, dataset):
masker = self.masker or dataset.masker
if not isinstance(masker, NiftiMasker):
raise ValueError(
f"A {type(masker)} mask has been detected. "
"Only NiftiMaskers are allowed for this Estimator. "
"This is because aggregation, such as averaging values across ROIs, "
"will produce invalid results."
)
pymare_dset = pymare.Dataset(y=self.inputs_["z_maps"])
est = pymare.estimators.FisherCombinationTest()
est.fit_dataset(pymare_dset)
est_summary = est.summary()
results = {
"z": boolean_unmask(est_summary.z.squeeze(), self.inputs_["aggressive_mask"]),
"p": boolean_unmask(est_summary.p.squeeze(), self.inputs_["aggressive_mask"]),
}
return results
(lower_bound, upper_bound),
(N_STUDIES + 1, N_STUDIES + 1),
color="green",
linewidth=3,
label="Between-Study 95% CI",
)
ax.set_ylim((0, N_STUDIES + 2))
ax.set_xlabel("Mean (95% CI)")
ax.set_ylabel("Study")
ax.legend()
fig.tight_layout()
###############################################################################
# Create a Dataset object containing the data
# --------------------------------------------
dset = pymare.Dataset(y=y, X=None, v=v, n=n, add_intercept=True)
# Here is a dictionary to house results across models
results = {}
###############################################################################
# Fit models
# -----------------------------------------------------------------------------
# When you have ``z`` or ``p``:
#
# - :class:`pymare.estimators.Stouffers`
# - :class:`pymare.estimators.Fishers`
#
# When you have ``y`` and ``v`` and don't want to estimate between-study variance:
#
# - :class:`pymare.estimators.WeightedLeastSquares`
