def main():
"""
Entry point for analysis
"""
rsms = load_rsms()
rdms = {partition: 1 - x for (partition, x) in rsms.items()}
intersubject_corr_dict = {}
for partition_name, partition_rdms in rdms.items():
lower_tris = np.array([
get_lower_tri(subject_rdm, with_diagonal=True)
for subject_rdm in partition_rdms
])
lower_tri_corrs = np.corrcoef(lower_tris)
# reduce symmetric corrmat by taking lower triangle (without diagonal)
pairwise_intersubject_corrs = get_lower_tri(lower_tri_corrs)
intersubject_corr_dict[partition_name] = pairwise_intersubject_corrs
print(f"Mean intersub RDM corr in {partition_name}")
report_ttest_1_sample(pairwise_intersubject_corrs, print_mean_var=True)
print(f"Lateral vs. medial difference")
report_ttest_2_sample(intersubject_corr_dict["VTC_lateral"],
intersubject_corr_dict["VTC_medial"])
def get_partition_weights(
thresh="thr_75",
glm_dir="GLM_vanilla",
metric=PRIMARY_METRIC,
r2_control=False,
noise_type="none",
):
rsms = load_rsms(
thresh=thresh,
glm_dir=glm_dir,
r2_control=r2_control,
metric=metric,
noise_type=noise_type,
)
rdms = {partition: 1 - x for (partition, x) in rsms.items()}
regressors = build_regressors()
# fit domain, category, and depth weights to each rdm
partition_weights = {}
for partition in PARTITIONS:
partition_rdms = rdms[partition]
weights_list = []
for rdm in partition_rdms:
lower_tri_rdm = get_lower_tri(rdm, with_diagonal=True)
weights = fit_rdm(lower_tri_rdm, regressors, method="lsq")
weights_list.append(weights)
partition_weights[partition] = np.stack(weights_list)
return partition_weights
def get_partition_weights(
thresh="thr_75", glm_dir="GLM_vanilla", r2_control=False, noise_type="none"
):
rsms = load_rsms(
thresh=thresh, glm_dir=glm_dir, r2_control=r2_control, noise_type=noise_type
)
rdms_by_depth = {
partition: np.array([1 - split_by_depth(rsm) for rsm in partition_rsms])
for (partition, partition_rsms) in rsms.items()
}
regressors = build_regressors()
partition_weights = {}
for partition in PARTITIONS:
partition_rdms = rdms_by_depth[partition]
subject_weights = []
for subject_rdms in partition_rdms:
depth_weights = []
for depth_rdm in subject_rdms:
weights = fit_rdm(
get_lower_tri(depth_rdm, with_diagonal=True),
regressors,
method="lsq",
)
depth_weights.append(weights)
subject_weights.append(np.stack(depth_weights))
partition_weights[partition] = np.stack(subject_weights)
return partition_weights
def load_rdms():
"""
Retrieve distance matrices for "true" (3T 2.4mm) and "simulated" (7T downsampled
2.4mm) data
"""
sessions = {
"true": {
"fsid_session": "C1051_20161006",
"glm_dir": "GLM_vanilla"
},
"simulated": {
"fsid_session": "C1051_20160212",
"glm_dir": "GLM_vanilla_sim2pt4",
},
}
rdms = {}
for session_name, session_info in sessions.items():
rdms[session_name] = {}
for partition in PARTITIONS:
rsm = load_single_rsm(
PATHS["rsms"],
session_info["fsid_session"],
glm_dir=session_info["glm_dir"],
partition=partition,
)
rdms[session_name][partition] = get_lower_tri(1 - rsm,
with_diagonal=True)
return rdms
def build_regressors(add_intercept=True):
"""
Provides regressor matrix for linear fits
"""
regressor_indices = [
np.repeat(np.arange(5), 2), # domains
np.arange(10), # categories
]
regressors = []
for indices in regressor_indices:
base = np.zeros((10, 10))
xx, yy = np.meshgrid(indices, indices)
base[xx == yy] = 1
regressors.append(get_lower_tri(1 - base, with_diagonal=True))
if add_intercept:
ones_matrix = np.ones((10, 10))
regressors.append(get_lower_tri(ones_matrix, with_diagonal=True))
return np.array(regressors)
def main():
"""
Entry point for analysis
"""
rsms = load_rsms()
rdms = {name: 1 - rsm for (name, rsm) in rsms.items()}
partition_pairs = {
"lateral-medial": ["VTC_lateral", "VTC_medial"],
"lateral-V1": ["VTC_lateral", "hOc1"],
"medial-V1": ["VTC_medial", "hOc1"],
}
pair_corrs = {}
for name, pair in partition_pairs.items():
rdms_a = rdms[pair[0]]
rdms_b = rdms[pair[1]]
subject_corrs = []
for a, b in zip(rdms_a, rdms_b):
lower_tri_a = get_lower_tri(a)
lower_tri_b = get_lower_tri(b)
corr = np.corrcoef(np.stack((lower_tri_a, lower_tri_b)))[0, 1]
subject_corrs.append(corr)
print(f"\nCorrelations between {pair[0]} and {pair[1]}")
pair_corrs[name] = np.array(subject_corrs)
report_ttest_1_sample(np.array(subject_corrs), print_mean_var=True)
print(f"\nt-test between lateral-medial and lateral-V1")
report_ttest_2_sample(pair_corrs["lateral-medial"],
pair_corrs["lateral-V1"])
print(f"\nt-test between lateral-medial and medial-V1")
report_ttest_2_sample(pair_corrs["lateral-medial"],
pair_corrs["medial-V1"])
def prepare_dataframe(hires_rdms, lowres_rdms):
"""
Sets up pandas dataframe with the following fields:
- partitions
- subjects
- data_srcs (resolutions)
- magnitudes
"""
data_src_list = list()
subjects_list = list()
partitions_list = list()
magnitudes_list = list()
for data_src_idx, rdms in enumerate([hires_rdms, lowres_rdms]):
for partition_idx, partition_name in enumerate(["VTC_lateral", "VTC_medial"]):
# extract mean of lower triangle of partition RDM, keeping diagonal
# because values are strictly positive, no need to take absolute value here
partition_rdms = rdms[partition_name]
lower_triangle_means = np.array(
[np.mean(get_lower_tri(x, with_diagonal=True)) for x in partition_rdms]
)
for sub_idx, val in enumerate(lower_triangle_means):
magnitudes_list.append(val)
data_src_list.append(data_src_idx)
subjects_list.append(sub_idx)
partitions_list.append(partition_idx)
subjects_list = np.array(subjects_list)
data_src_list = np.array(data_src_list)
partitions_list = np.array(partitions_list)
magnitudes_list = np.array(magnitudes_list)
data = {
"partitions": partitions_list,
"subjects": subjects_list,
"data_srcs": data_src_list,
"magnitudes": magnitudes_list,
}
return pd.DataFrame(data=data)
