Skip to content

Commit

Permalink
minor edits
Browse files Browse the repository at this point in the history
  • Loading branch information
@OnnoKampman
OnnoKampman committed Apr 22, 2024
1 parent 78fd31d commit 4b251e7
Show file tree
Hide file tree
Showing 11 changed files with 38 additions and 1,839 deletions.
4 changes: 3 additions & 1 deletion benchmarks/fmri/sim/imputation_study/compute_LEOO_likelihoods.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@
f'{covs_type:s}_covariance.csv'
)
if not os.path.exists(data_filepath):
logging.warning(f"File '{data_filepath:s}' not found.")
logging.warning(f"Data file '{data_filepath:s}' not found.")
if covs_type == 'boxcar':
data_filepath = os.path.join(
cfg['data-dir'], noise_type, f'trial_{i_trial:03d}',
Expand All @@ -48,6 +48,8 @@
if not os.path.exists(data_filepath):
logging.warning(f"File '{data_filepath:s}' not found.")
continue
else:
continue
x, y = load_data(
data_filepath,
verbose=False,
Expand Down
64 changes: 35 additions & 29 deletions helpers/evaluation.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -241,11 +241,11 @@ def get_tvfc_estimates(

repetition_time = None

wp_joint_model_savedir = os.path.join(
wp_model_savedir = os.path.join(
config_dict['experiments-basedir'], noise_type, data_split,
f'trial_{i_trial:03d}', model_name
)
wp_joint_model_filename = f'{covs_type:s}.json'
wp_model_filename = f'{covs_type:s}.json'

tvfc_estimates_savedir = os.path.join(
config_dict['experiments-basedir'], noise_type, f'trial_{i_trial:03d}',
Expand All @@ -256,10 +256,10 @@ def get_tvfc_estimates(
)

# Fix renaming issue.
if not os.path.exists(os.path.join(wp_joint_model_savedir, wp_joint_model_filename)):
logging.warning(f"Model file {os.path.join(wp_joint_model_savedir, wp_joint_model_filename):s} not found.")
if not os.path.exists(os.path.join(wp_model_savedir, wp_model_filename)):
logging.warning(f"WP model file {os.path.join(wp_model_savedir, wp_model_filename):s} not found.")
if covs_type == 'boxcar':
wp_joint_model_filename = 'checkerboard.json'
wp_model_filename = 'checkerboard.json'
if not os.path.exists(tvfc_estimates_filepath):
if covs_type == 'boxcar':
tvfc_estimates_filepath = os.path.join(
Expand All @@ -270,11 +270,11 @@ def get_tvfc_estimates(
assert experiment_dimensionality is not None
assert scan_id is not None
assert subject is not None
wp_joint_model_savedir = os.path.join(
wp_model_savedir = os.path.join(
config_dict['experiments-basedir'], 'saved_models', f'scan_{scan_id:d}',
data_split, experiment_dimensionality, model_name
)
wp_joint_model_filename = f'{subject:d}.json'
wp_model_filename = f'{subject:d}.json'
repetition_time = config_dict['repetition-time']
# tvfc_estimates_filepath = os.path.join(
# config_dict['experiments-basedir'], 'TVFC_estimates', f'scan_{scan_id:d}',
Expand All @@ -283,27 +283,27 @@ def get_tvfc_estimates(
case 'rockland':
pp_pipeline = 'custom_fsl_pipeline'
assert subject is not None
wp_joint_model_savedir = os.path.join(
wp_model_savedir = os.path.join(
config_dict['experiments-basedir'], pp_pipeline, 'saved_models', data_split, model_name
)
wp_joint_model_filename = f"{subject.removesuffix('.csv'):s}.json"
wp_model_filename = f"{subject.removesuffix('.csv'):s}.json"
repetition_time = config_dict['repetition-time']
case _:
logging.error(f"Dataset '{data_set_name:s}' not recognized.")
return

match model_name:
case 'VWP' | 'VWP_joint':
if os.path.exists(os.path.join(wp_joint_model_savedir, wp_joint_model_filename)):
if os.path.exists(os.path.join(wp_model_savedir, wp_model_filename)):
k = gpflow.kernels.Matern52()
m = VariationalWishartProcess(
x_train, y_train,
nu=n_time_series,
kernel=k
)
m.load_from_params_dict(
savedir=wp_joint_model_savedir,
model_name=wp_joint_model_filename,
savedir=wp_model_savedir,
model_name=wp_model_filename,
)
if metric == 'correlation':
all_covs_means, _ = m.predict_corr(x_train) # Tensor of shape (N, D, D), _
Expand All @@ -312,10 +312,10 @@ def get_tvfc_estimates(
estimated_tvfc = all_covs_means.numpy() # (N, D, D)
del m
else:
logging.warning(f"VWP model not found in '{wp_joint_model_savedir:s}'.")
logging.warning(f"VWP model not found in '{wp_model_savedir:s}'.")
return
case 'SVWP' | 'SVWP_joint':
if os.path.exists(os.path.join(wp_joint_model_savedir, wp_joint_model_filename)):
if os.path.exists(os.path.join(wp_model_savedir, wp_model_filename)):
k = gpflow.kernels.Matern52()
m = SparseVariationalWishartProcess(
D=n_time_series,
Expand All @@ -325,8 +325,8 @@ def get_tvfc_estimates(
verbose=False
)
m.load_from_params_dict(
savedir=wp_joint_model_savedir,
model_name=wp_joint_model_filename,
savedir=wp_model_savedir,
model_name=wp_model_filename,
)
if metric == 'correlation':
all_covs_means, _ = m.predict_corr(x_train) # Tensor of shape (N, D, D), _
Expand All @@ -335,7 +335,7 @@ def get_tvfc_estimates(
estimated_tvfc = all_covs_means.numpy() # (N, D, D)
del m
else:
logging.warning(f"SVWP model not found in '{wp_joint_model_savedir:s}'.")
logging.warning(f"SVWP model not found in '{wp_model_savedir:s}'.")
return
case 'DCC' | 'DCC_joint' | 'DCC_bivariate_loop' | 'GO' | 'GO_joint' | 'GO_bivariate_loop':
if os.path.exists(tvfc_estimates_filepath):
Expand Down Expand Up @@ -405,20 +405,26 @@ def get_test_location_estimated_covariance_structure(
subject: str,
data_split: str,
y_train_locations: np.array = None,
scan_id: int = None, experiment_dimensionality: str = None,
scan_id: int = None,
experiment_dimensionality: str = None,
noise_type: str = None,
i_trial: int = None,
covs_type: str = None,
connectivity_metric: str = 'covariance',
):
) -> np.array:
"""
Estimates test location covariance structure.
For all methods except the Wishart process models, we load the pre-saved train locations predictions.
Parameters
----------
experiment_dimensionality: str
'multivariate' or 'bivariate'.
"""
data_set_name = config_dict['data-set-name']
match data_set_name:
case 'd2' | 'd3d' | 'd3s' | 'd4s' | 'd6s' | 'd9s' | 'd15s': # TODO: generalize this
model_savedir = os.path.join(
wp_model_savedir = os.path.join(
config_dict['experiments-basedir'], noise_type, data_split,
f'trial_{i_trial:03d}', model_name
)
Expand All @@ -433,8 +439,8 @@ def get_test_location_estimated_covariance_structure(
)

# Fix renaming issue.
if not os.path.exists(os.path.join(model_savedir, wp_model_filename)):
logging.warning(f"Model file {os.path.join(model_savedir, wp_model_filename):s} not found.")
if not os.path.exists(os.path.join(wp_model_savedir, wp_model_filename)):
logging.warning(f"WP model file {os.path.join(wp_model_savedir, wp_model_filename):s} not found.")
if covs_type == 'boxcar':
wp_model_filename = 'checkerboard.json'
if not os.path.exists(tvfc_estimates_filepath):
Expand All @@ -446,7 +452,7 @@ def get_test_location_estimated_covariance_structure(
case 'HCP_PTN1200_recon2':
assert scan_id is not None
assert experiment_dimensionality is not None
model_savedir = os.path.join(
wp_model_savedir = os.path.join(
config_dict['experiments-basedir'], 'saved_models', f'scan_{scan_id:d}',
data_split, experiment_dimensionality, model_name
)
Expand All @@ -458,7 +464,7 @@ def get_test_location_estimated_covariance_structure(
)
case 'rockland':
pp_pipeline = 'custom_fsl_pipeline'
model_savedir = os.path.join(
wp_model_savedir = os.path.join(
config_dict['experiments-basedir'], pp_pipeline, 'saved_models', data_split, model_name
)
wp_model_filename = f"{subject.removesuffix('.csv'):s}.json"
Expand All @@ -472,25 +478,25 @@ def get_test_location_estimated_covariance_structure(

match model_name:
case 'VWP' | 'VWP_joint': # we do not have to interpolate linearly here
wp_model_filepath = os.path.join(model_savedir, wp_model_filename)
wp_model_filepath = os.path.join(wp_model_savedir, wp_model_filename)
if not os.path.exists(wp_model_filepath):
raise FileNotFoundError(f"Could not load model '{wp_model_filepath:s}'.")
raise FileNotFoundError(f"Could not load WP model '{wp_model_filepath:s}'.")
k = gpflow.kernels.Matern52()
m = VariationalWishartProcess(
x_observed=x_train_locations,
y_observed=y_train_locations,
nu=n_time_series,
kernel=k
)
m.load_from_params_dict(model_savedir, wp_model_filename)
m.load_from_params_dict(wp_model_savedir, wp_model_filename)
if connectivity_metric == 'correlation':
all_covs_means, _ = m.predict_corr(x_test_locations) # Tensor of shape (N, D, D), _
else:
all_covs_means, _ = m.predict_cov(x_test_locations) # Tensor of shape (N, D, D), _
test_locations_predicted_covariance_structure = all_covs_means.numpy() # (N, D, D)
del m
case 'SVWP' | 'SVWP_joint': # we do not have to interpolate linearly here
wp_model_filepath = os.path.join(model_savedir, wp_model_filename)
wp_model_filepath = os.path.join(wp_model_savedir, wp_model_filename)
if not os.path.exists(wp_model_filepath):
raise FileNotFoundError(f"Could not load model '{wp_model_filepath:s}'.")
k = gpflow.kernels.Matern52()
Expand All @@ -502,7 +508,7 @@ def get_test_location_estimated_covariance_structure(
verbose=False
)
m.load_from_params_dict(
savedir=model_savedir,
savedir=wp_model_savedir,
model_name=wp_model_filename,
)
test_locations_predicted_covariance_structure, _ = m.predict_cov(x_new=x_test_locations) # Tensor of shape (N_test, D, D)
Expand Down
Loading

0 comments on commit 4b251e7

Please sign in to comment.