brainscore-vision 2.1__py3-none-any.whl

brainscore_vision/data/cadena2017/data_packaging/analysis.py

@@ -0,0 +1,26 @@
+import numpy as np
+from PIL import Image
+import os
+
+
+def main():
+    data_path = os.path.join(os.path.dirname(__file__), 'cadena2017')
+    images = np.load(os.path.join(data_path, 'images.npy'))
+    _show_image(images[0])
+    np.testing.assert_array_equal(images.shape, (7250, 140, 140))  # images x width x height
+    responses = np.load(os.path.join(data_path, 'responses.npy'))
+    np.testing.assert_array_equal(responses.shape, (4, 7250, 166))  # num_repetitions x num_images x num_neurons
+    print("{}/{} responses NaN ({:.2f}%)".format(
+        np.isnan(responses).sum(), responses.size, np.isnan(responses).sum() / responses.size * 100))
+
+
+def _show_image(img, savepath=None):
+    img = Image.fromarray((img * 255).astype('uint8'))
+    if savepath:
+        img.save(savepath)
+    else:
+        img.show()
+
+
+if __name__ == '__main__':
+    main()

brainscore_vision/data/cadena2017/test.py

@@ -0,0 +1,24 @@
+import pytest
+
+from brainscore_vision import load_dataset, load_stimulus_set
+from brainscore_vision.benchmark_helpers import check_standard_format
+
+
+@pytest.mark.private_access
+def test_format():
+    assembly = load_dataset('Cadena2017')
+    check_standard_format(assembly, nans_expected=True)
+
+
+@pytest.mark.private_access
+def test_num_items():
+    assembly = load_dataset('Cadena2017')
+    assert assembly.attrs['stimulus_set_identifier'] == 'Cadena2017'
+    assert len(assembly['presentation']) == 29000
+    assert len(assembly['neuroid']) == 166
+
+
+@pytest.mark.private_access
+def test_stimulus_set():
+    stimulus_set = load_stimulus_set('Cadena2017')
+    assert len(stimulus_set) == 7249

brainscore_vision/data/cichy2019/__init__.py

@@ -0,0 +1,38 @@
+from brainio.assemblies import DataAssembly
+from brainscore_vision import data_registry, stimulus_set_registry, load_stimulus_set
+from brainscore_vision.data_helpers.s3 import load_assembly_from_s3, load_stimulus_set_from_s3
+
+BIBTEX = """@article{CICHY201912,
+title = {The spatiotemporal neural dynamics underlying perceived similarity for real-world objects},
+journal = {NeuroImage},
+volume = {194},
+pages = {12-24},
+year = {2019},
+issn = {1053-8119},
+doi = {https://doi.org/10.1016/j.neuroimage.2019.03.031},
+url = {https://www.sciencedirect.com/science/article/pii/S1053811919302083},
+author = {Radoslaw M. Cichy and Nikolaus Kriegeskorte and Kamila M. Jozwik and Jasper J.F. {van den Bosch} and Ian Charest},
+keywords = {fMRI, MEG, Object recognition, Perceived similarity, Visual perception},
+abstract = {The degree to which we perceive real-world objects as similar or dissimilar structures our perception and guides categorization behavior. Here, we investigated the neural representations enabling perceived similarity using behavioral judgments, fMRI and MEG. As different object dimensions co-occur and partly correlate, to understand the relationship between perceived similarity and brain activity it is necessary to assess the unique role of multiple object dimensions. We thus behaviorally assessed perceived object similarity in relation to shape, function, color and background. We then used representational similarity analyses to relate these behavioral judgments to brain activity. We observed a link between each object dimension and representations in visual cortex. These representations emerged rapidly within 200 ms of stimulus onset. Assessing the unique role of each object dimension revealed partly overlapping and distributed representations: while color-related representations distinctly preceded shape-related representations both in the processing hierarchy of the ventral visual pathway and in time, several dimensions were linked to high-level ventral visual cortex. Further analysis singled out the shape dimension as neither fully accounted for by supra-category membership, nor a deep neural network trained on object categorization. Together our results comprehensively characterize the relationship between perceived similarity of key object dimensions and neural activity.}
+}"""
+
+
+# assembly
+data_registry['Cichy2019'] = lambda: load_assembly_from_s3(
+    identifier="aru.Cichy2019",
+    version_id="83yFgsx5rc6pgpra4UiMxDDvgaV6ytU8",
+    sha1="701e63be62b642082d476244d0d91d510b3ff05d",
+    bucket="brainio-brainscore",
+    cls=DataAssembly,
+    stimulus_set_loader=lambda: load_stimulus_set('Cichy2019'),
+)
+
+
+# stimulus set
+stimulus_set_registry['Cichy2019'] = lambda: load_stimulus_set_from_s3(
+    identifier="Cichy2019",
+    bucket="brainio-brainscore",
+    csv_sha1="281c4d9d0dd91a2916674638098fe94afb87d29a",
+    zip_sha1="d2166dd9c2720cb24bc520f5041e6830779c0240",
+    csv_version_id="NXC3IR_XZG9QuMtxL7QFtDIcmGJrMKz4",
+    zip_version_id="5lAaoRzViKRDRyqiTYXm2PB3ADaV2Szo")

brainscore_vision/data/cichy2019/test.py

@@ -0,0 +1,8 @@
+import pytest
+
+from brainscore_vision import load_dataset
+
+
+@pytest.mark.private_access
+def test_existence():
+    assert load_dataset('Cichy2019') is not None

brainscore_vision/data/coggan2024_behavior/__init__.py

@@ -0,0 +1,36 @@
+# Created by David Coggan on 2024 06 23
+
+from brainio.assemblies import BehavioralAssembly
+from brainscore_vision import (
+    stimulus_set_registry, data_registry, load_stimulus_set)
+from brainscore_vision.data_helpers.s3 import (
+    load_assembly_from_s3, load_stimulus_set_from_s3)
+
+# stimulus set
+stimulus_set_registry['Coggan2024_behavior'] = lambda: load_stimulus_set_from_s3(
+    identifier="tong.Coggan2024_behavior",
+    bucket="brainio-brainscore",
+    csv_sha1="01c312c4c16f7acc5afddaafcf826e5af58b13e2",
+    zip_sha1="1c070b88fa45e9e69d58f95466cb6406a45a4873",
+    csv_version_id="oyBtiZLEnNnCOHGzH5e12uzO9FrQ.ja_",
+    zip_version_id="sgPkAllzqOgP6KZcuGenFdAecBmRYApo")
+
+# fitting stimuli
+stimulus_set_registry['Coggan2024_behavior_fitting'] = lambda: (
+    load_stimulus_set_from_s3(
+        identifier="tong.Coggan2024_behavior_fitting",
+        bucket="brainio-brainscore",
+        csv_sha1="136e48992305ea78a4fb77e9dfc75dcf01e885d0",
+        zip_sha1="24e68f5ba2f8f2105daf706307642637118e7d36",
+        csv_version_id="wq1GTpsTqB9ODddVSWpd4HMOllY7Q08_",
+        zip_version_id="IU.vvRmzBBhxErneMgQHoB1OKaeyen17"))
+
+# behavioral data
+data_registry['Coggan2024_behavior'] = lambda: load_assembly_from_s3(
+    identifier="tong.Coggan2024_behavior",
+    version_id="Xuov.DxZnqSD1.Ia.ZySbSu8IW7cFigT",
+    sha1="c1ac4a268476c35bbe40081358667a03d3544631",
+    bucket="brainio-brainscore",
+    cls=BehavioralAssembly,
+    stimulus_set_loader=lambda: load_stimulus_set('Coggan2024_behavior'),
+)

brainscore_vision/data/coggan2024_behavior/data_packaging.py

@@ -0,0 +1,166 @@
+# Created by David Coggan on 2024 06 23
+
+# package stimuli
+from brainio.stimuli import StimulusSet
+from brainio.packaging import package_stimulus_set, package_data_assembly
+from brainio.assemblies import BehavioralAssembly
+import os
+import os.path as op
+import pandas as pd
+import numpy as np
+import torchvision.transforms as transforms
+from PIL import Image
+
+
+imagenet_metadata = dict(
+    bear=dict(
+        class_index_1K=294,
+        class_name='n02132136',
+        class_description='brown bear, bruin, Ursus arctos'),
+    bison=dict(
+        class_index_1K=347,
+        class_name='n02410509',
+        class_description='bison'),
+    elephant=dict(
+        class_index_1K=386,
+        class_name='n02504458',
+        class_description='African elephant, Loxodonta africana'),
+    hare=dict(
+        class_index_1K=331,
+        class_name='n02326432',
+        class_description='hare'),
+    jeep=dict(
+        class_index_1K=609,
+        class_name='n03594945',
+        class_description='jeep, landrover'),
+    lamp=dict(
+        class_index_1K=846,
+        class_name='n04380533',
+        class_description='table lamp'),
+    sportsCar=dict(
+        class_index_1K=817,
+        class_name='n04285008',
+        class_description='sports car, sport car'),
+    teapot=dict(
+        class_index_1K=849,
+        class_name='n04398044',
+        class_description='teapot')
+)
+
+
+"""
+Stimuli used in human behavioral experiment. Each subject was shown a unique 
+set of 753 occluded images so the entire stimulus set is 22560 images.
+"""
+trials = pd.read_parquet('/home/tonglab/david/projects/p022_occlusion/in_vivo'
+                         '/behavioral/exp1/analysis/trials.parquet')
+trials.object_class.replace('car', 'sportsCar', inplace=True)
+trials.prediction.replace('car', 'sportsCar', inplace=True)
+stimuli = []  # collect meta
+stimulus_paths = {}
+exp_dir = (
+    '/home/tonglab/david/projects/p022_occlusion/in_vivo/behavioral/exp1/data')
+subjects = sorted(trials.subject.unique())
+for t, trial in trials.iterrows():
+    subj = subjects.index(trial.subject)  # 0-indexed
+    stim_path = exp_dir + trial.occluded_object_path.split('logFiles')[1]
+    stimulus_id = f'{t:05}_sub-{subj:02}_trial-{trial.trial-1:03}'
+    object_class, occluder_type, coverage, occluder_color, rep = (
+        op.basename(stim_path).split('.png')[0].split('_'))
+    visibility = np.round(1 - float(coverage), 1)
+    object_data = imagenet_metadata[object_class]
+    stimulus_paths[stimulus_id] = stim_path
+    stimuli.append({
+        'stimulus_id': stimulus_id,
+        'subject': f'sub-{subj:02}',
+        'trial': trial.trial-1,
+        'object_class': object_class,
+        'imagenet_class_index_1K': object_data['class_index_1K'],
+        'imagenet_class_name': object_data['class_name'],
+        'imagenet_class_description': object_data['class_description'],
+        'occluder_type': occluder_type,
+        'occluder_color': occluder_color,
+        'visibility': visibility,
+        'repetition': int(rep)})
+stimuli = StimulusSet(stimuli)
+stimuli.stimulus_paths = stimulus_paths
+stimuli.name = "tong.Coggan2024_behavior"
+
+assert len(stimuli) == 22560
+
+packaged_stimulus_metadata = package_stimulus_set(
+    catalog_name=None,
+    proto_stimulus_set=stimuli,
+    stimulus_set_identifier=stimuli.name,
+    bucket_name="brainio-brainscore")
+print(packaged_stimulus_metadata)
+
+
+"""
+Stimuli used to fit the models.
+These are an independent set of imagenet images from the same 8 classes as the 
+behavioral experiment.
+"""
+# get stimuli pngs
+transform = transforms.Compose([
+    transforms.ToTensor(),
+    transforms.Resize(224),
+    transforms.CenterCrop(224),
+    transforms.Grayscale(num_output_channels=3),
+    transforms.ToPILImage()])
+svc_dataset = pd.read_csv('/home/tonglab/david/projects/p022_occlusion/'
+    'in_silico/analysis/scripts/utils/SVC_images.csv')
+svc_dataset['class'].replace('car', 'sportsCar', inplace=True)
+image_dir = ('/home/tonglab/david/projects/p022_occlusion/in_silico/images'
+             '/behavior_svc')
+#os.makedirs(image_dir, exist_ok=True)
+#for image_path in svc_dataset.filepath.values:
+#    image = Image.open(image_path)
+#    image = transform(image)
+#    image.save(f'{image_dir}/{op.basename(image_path).split(".")[0]}.png')
+
+# package stimuli
+stimuli = []
+stimulus_paths = {}
+for i, row in svc_dataset.iterrows():
+    stimulus_id = f'{i:04}_{op.basename(row.filepath).split(".")[0]}'
+    stimulus_paths[stimulus_id] = \
+        f'{image_dir}/{op.basename(row.filepath).split(".")[0]}.png'
+    stimuli.append({
+        'stimulus_id': stimulus_id,
+        'image_label': row['class'],
+        'imagenet_path': row.filepath.split('ILSVRC2012/')[-1]})
+stimuli = StimulusSet(stimuli)
+stimuli.stimulus_paths = stimulus_paths
+stimuli.name = "tong.Coggan2024_behavior_fitting"
+assert len(stimuli) == 2048
+packaged_stimulus_metadata = package_stimulus_set(
+    catalog_name=None,
+    proto_stimulus_set=stimuli,
+    stimulus_set_identifier=stimuli.name,
+    bucket_name="brainio-brainscore")
+print(packaged_stimulus_metadata)
+
+
+"""
+Human behavioral responses to occluded images.
+"""
+# package data assembly
+predictions = trials.prediction.to_list()
+stimulus_ids = stimuli.stimulus_id.to_list()
+assembly = BehavioralAssembly(
+    predictions,
+    dims=['presentation'],
+    coords={'stimulus_id': ('presentation', stimulus_ids),
+            'human_accuracy': ('presentation', trials.accuracy)})
+assembly.name = "tong.Coggan2024_behavior"
+
+# upload to S3
+packaged_behavioral_metadata = package_data_assembly(
+    proto_data_assembly=assembly,
+    assembly_identifier=assembly.name,
+    stimulus_set_identifier=stimuli.name,
+    assembly_class_name="BehavioralAssembly",
+    bucket_name="brainio-brainscore",
+    catalog_identifier=None)
+print(packaged_behavioral_metadata)

brainscore_vision/data/coggan2024_behavior/test.py

@@ -0,0 +1,32 @@
+# Created by David Coggan on 2024 06 26
+
+import pytest
+import numpy as np
+import brainscore_vision
+
+@pytest.mark.private_access
+def test_Coggan2024_behavior_stimuli():
+    stimulus_set = brainscore_vision.load_stimulus_set('Coggan2024_behavior')
+    assert len(stimulus_set) == 22560
+    assert len(set(stimulus_set['object_class'])) == 8
+    assert len(set(stimulus_set['occluder_type'])) == 10
+    assert len(set(stimulus_set['occluder_color'])) == 3
+    assert len(set(stimulus_set['visibility'])) == 6
+
+@pytest.mark.private_access
+def test_Coggan2024_behavior_stimuli_fitting():
+    stimulus_set = brainscore_vision.load_stimulus_set(
+        'Coggan2024_behavior_fitting')
+    assert len(stimulus_set) == 2048
+
+@pytest.mark.private_access
+def test_Coggan2024_behavior_dataset():
+    assembly = brainscore_vision.load_dataset('Coggan2024_behavior')
+    np.testing.assert_array_equal(
+        assembly.dims, ['presentation'])
+    assert len(set(assembly['stimulus_id'].values)) == 22560
+    assert assembly.shape[0] == 22560
+    assert assembly.stimulus_set is not None
+    assert len(assembly.stimulus_set) == 22560
+
+

brainscore_vision/data/coggan2024_fMRI/__init__.py

@@ -0,0 +1,27 @@
+# Created by David Coggan on 2024 06 23
+
+from brainio.assemblies import NeuroidAssembly
+from brainscore_vision import (
+    stimulus_set_registry, data_registry, load_stimulus_set)
+from brainscore_vision.data_helpers.s3 import (
+    load_assembly_from_s3, load_stimulus_set_from_s3)
+
+# stimulus set
+stimulus_set_registry['Coggan2024_fMRI'] = lambda: load_stimulus_set_from_s3(
+    identifier="tong.Coggan2024_fMRI",
+    bucket="brainio-brainscore",
+    csv_sha1="0089f5f8fd3f2de14de12ed736a0f88575f8e1ee",
+    zip_sha1="e26fdea4d866799526dea183f5bfb9792718822a",
+    csv_version_id="q0kxLCC8m6LSaLEQzB26sfyoKgJJwkAs",
+    zip_version_id="SZwyN4ZgEmpW22YlcbqIbL9c0QudlMAp")
+
+# fMRI data
+data_registry['Coggan2024_fMRI'] = lambda: load_assembly_from_s3(
+    identifier="tong.Coggan2024_fMRI",
+    version_id="6P898Mio3VBsFx_qbC9rJECMELNGRprH",
+    sha1="da3adbca5247d0491d366f94e8431fb3e4e58db2",
+    bucket="brainio-brainscore",
+    cls=NeuroidAssembly,
+    stimulus_set_loader=lambda: load_stimulus_set('Coggan2024_fMRI'),
+)
+

brainscore_vision/data/coggan2024_fMRI/data_packaging.py

@@ -0,0 +1,123 @@
+# Created by David Coggan on 2024 06 23
+
+from brainio.stimuli import StimulusSet
+from brainio.packaging import package_stimulus_set
+import glob
+import os.path as op
+from brainio.assemblies import DataAssembly, NeuroidAssembly
+from brainio.packaging import package_data_assembly
+import pickle as pkl
+from itertools import product as itp
+import numpy as np
+
+
+# imagenet metadata
+imagenet_metadata = dict(
+    bear=dict(
+        class_index_1K=294,
+        class_name='n02132136',
+        class_description='brown bear, bruin, Ursus arctos',
+        path='val/n02132136/ILSVRC2012_val_00049345.jpg'),
+    bison=dict(
+        class_index_1K=347,
+        class_name='n02410509',
+        class_description='bison',
+        path='val/n02410509/ILSVRC2012_val_00048511.jpg'),
+    elephant=dict(
+        class_index_1K=386,
+        class_name='n02504458',
+        class_description='African elephant, Loxodonta africana',
+        path='val/n02504458/ILSVRC2012_val_00030840.jpg'),
+    hare=dict(
+        class_index_1K=331,
+        class_name='n02326432',
+        class_description='hare',
+        path='val/n02326432/ILSVRC2012_val_00004064.jpg'),
+    jeep=dict(
+        class_index_1K=609,
+        class_name='n03594945',
+        class_description='jeep, landrover',
+        path='val/n03594945/ILSVRC2012_val_00036304.jpg'),
+    lamp=dict(
+        class_index_1K=846,
+        class_name='n04380533',
+        class_description='table lamp',
+        path='val/n04380533/ILSVRC2012_val_00001055.jpg'),
+    sportsCar=dict(
+        class_index_1K=817,
+        class_name='n04285008',
+        class_description='sports car, sport car',
+        path='val/n04285008/ILSVRC2012_val_00001247.jpg'),
+    teapot=dict(
+        class_index_1K=849,
+        class_name='n04398044',
+        class_description='teapot',
+        path='val/n04398044/ILSVRC2012_val_00033663.jpg')
+)
+
+# stimuli
+stimuli = []  # collect meta
+stimulus_paths = {}  # collect mapping of stimulus_id to filepath
+for f, filepath in enumerate(sorted(glob.glob('stimuli/*.png'))):
+    stimulus_id = op.basename(filepath).split('.')[0]
+    object_name, occlusion_condition = stimulus_id.split('_')
+    occlusion_condition = occlusion_condition.split('.')[0]
+    stimulus_paths[stimulus_id] = filepath
+    object_data = imagenet_metadata[object_name]
+    stimuli.append({
+        'stimulus_id': stimulus_id,
+        'object_name': object_name,
+        'occlusion_condition': occlusion_condition,
+        'imagenet_class_index_1K': object_data['class_index_1K'],
+        'imagenet_class_name': object_data['class_name'],
+        'imagenet_class_description': object_data['class_description'],
+        'imagenet_path': object_data['path'],
+    })
+stimuli = StimulusSet(stimuli)
+stimuli.stimulus_paths = stimulus_paths
+stimuli.name = 'coggan2024_fMRI'
+"""
+packaged_stimulus_metadata = package_stimulus_set(
+    catalog_name=None,
+    proto_stimulus_set=stimuli,
+    stimulus_set_identifier=stimuli.name,
+    bucket_name="brainio-brainscore")
+pkl.dump(packaged_stimulus_metadata, open('packaged_stimulus_metadata.pkl', 'wb'))
+print(packaged_stimulus_metadata)
+"""
+
+
+# fMRI data
+dataset = pkl.load(open('dataset.pkl', 'rb'))
+regions = ['V1', 'V2', 'V4', 'IT']
+subjects = list(dataset['V1'].keys())
+n_subs = len(subjects)
+off_diag_indices = np.array(1 - np.eye(24).flatten(), dtype=bool)
+data_all, splits_all, sides_all, subjects_all, regions_all = [], [], [], [], []
+for region, subject in itp(regions, subjects):
+    data_all.append(dataset[region][subject])
+    subjects_all.append(subject)
+    regions_all.append(region)
+data_all = np.stack(data_all, axis=2)
+assembly = NeuroidAssembly(
+    data_all, dims=['presentation', 'presentation', 'neuroid'],
+    coords={
+        'stimulus_id': ('presentation', stimuli.stimulus_id),
+        'object_name': ('presentation', stimuli.object_name),
+        'occlusion_condition': ('presentation', stimuli.occlusion_condition),
+        'subject': ('neuroid', subjects_all),
+        'region': ('neuroid', regions_all),
+})
+assembly.name = 'coggan2024_fMRI'
+
+packaged_neural_metadata = package_data_assembly(
+    proto_data_assembly=assembly,
+    assembly_identifier=assembly.name,
+    stimulus_set_identifier=stimuli.name,
+    assembly_class_name="NeuroidAssembly",
+    bucket_name="brainio-brainscore",
+    catalog_identifier=None)
+
+# save the packaged metadata
+pkl.dump(packaged_neural_metadata, open('packaged_neural_metadata.pkl', 'wb'))
+print(packaged_neural_metadata)

brainscore_vision/data/coggan2024_fMRI/test.py

@@ -0,0 +1,25 @@
+# Created by David Coggan on 2024 06 26
+
+import brainscore_vision
+import numpy as np
+
+
+def test_Coggan2024_fMRI_stimuli():
+    stimulus_set = brainscore_vision.load_stimulus_set('Coggan2024_fMRI')
+    assert len(stimulus_set) == 24
+    assert len(set(stimulus_set['object_name'])) == 8
+    assert len(set(stimulus_set['occlusion_condition'])) == 3
+
+
+def test_Coggan2024_fMRI_dataset():
+    assembly = brainscore_vision.load_dataset('Coggan2024_fMRI')
+    np.testing.assert_array_equal(
+        assembly.dims, ['presentation', 'presentation', 'neuroid'])
+    assert len(set(assembly['stimulus_id'].values)) == 24
+    assert assembly.shape[0] == 24
+    assert assembly.shape[1] == 24
+    assert assembly.shape[2] == 36
+    assert assembly.stimulus_set is not None
+    assert len(assembly.stimulus_set) == 24
+
+

brainscore_vision/data/david2004/__init__.py

@@ -0,0 +1,34 @@
+from brainio.assemblies import NeuronRecordingAssembly
+
+from brainscore_vision import data_registry, stimulus_set_registry, load_stimulus_set
+from brainscore_vision.data_helpers.s3 import load_assembly_from_s3, load_stimulus_set_from_s3
+
+BIBTEX = """@article{david2004evaluation,
+  title={Evaluation of different measures of functional connectivity using a neural mass model},
+  author={David, Olivier and Cosmelli, Diego and Friston, Karl J},
+  journal={Neuroimage},
+  volume={21},
+  number={2},
+  pages={659--673},
+  year={2004},
+  publisher={Elsevier}
+}"""
+
+# assembly
+data_registry['David2004'] = lambda: load_assembly_from_s3(
+    identifier="gallant.David2004",
+    version_id="8getDVrrr1iT0DA385T8ZdSzCcuM3_m0",
+    sha1="d2ed9834c054da2333f5d894285c9841a1f27313",
+    bucket="brainio-brainscore",
+    cls=NeuronRecordingAssembly,
+    stimulus_set_loader=lambda: load_stimulus_set('David2004'),
+)
+
+# stimulus set
+stimulus_set_registry['David2004'] = lambda: load_stimulus_set_from_s3(
+    identifier="David2004",
+    bucket="brainio-brainscore",
+    csv_sha1="8ec76338b998cadcdf1e57edd2dd992e2ab2355b",
+    zip_sha1="0200421d66a0613946d39cab64c00b561160016e",
+    csv_version_id="79aMj3kby7rS8FteIeL4g1jeyrEZojyH",
+    zip_version_id="A8sf8YmSJoDbjEvCvnkePF9APDPfAgw4")