brainscore-vision 2.1__py3-none-any.whl

brainscore_vision/data/igustibagus2024/merged_assembly/create_merged_assembly_and_stim.py

@@ -0,0 +1,152 @@
+# import libraries
+
+# general
+import numpy as np
+import pandas as pd
+import xarray as xr
+import tqdm as tqdm
+
+
+# brain-score specific
+import brainscore
+import brainio 
+from brainscore.benchmarks._neural_common import average_repetition
+from brainio.assemblies import NeuroidAssembly
+from brainio.packaging import write_netcdf # use this function to save it locally
+from brainio.packaging import package_data_assembly # use this function to push to S3
+from brainio.stimuli import StimulusSet
+from brainio.packaging import package_stimulus_set
+
+imgs_dir_path = '../images'
+dependencies_dir_path = '../dependencies'
+
+# oleo is already on S3
+assembly_oleo = brainscore.get_assembly('dicarlo.Sanghavi2021.domain_transfer')
+
+# pico is not on S3 yet, we load it from local 
+file_path = dependencies_dir_path + '/data_pico/assy_dicarlo_pico_domain_transfer.nc'
+assembly_pico = brainio.assemblies.DataAssembly.from_files(file_path)
+
+# stimulus set with background id
+csv_path = 'merged_stimulus_set.csv'
+merged_stimulus_set = pd.read_csv(csv_path)
+
+# need to add background id to the presentation dimension of oleo assembly
+assembly_oleo['background_id'] = ('presentation', 
+                                  np.array([merged_stimulus_set.background_id[merged_stimulus_set.stimulus_id == stim_id].values[0] 
+                                        for stim_id in assembly_oleo['stimulus_id'].values]))
+
+# small renaming and reordering of presentation and neuroid for assembly pico, to make it compatible with oleo
+# we also add the backgroud_id to the presentation
+data = assembly_pico.values
+coords = {
+    # selection of presentaiton
+    'object_label': ('presentation', assembly_pico['object_label'].values),
+    'object_style': ('presentation', assembly_pico['object_style'].values),
+    'filepath': ('presentation', assembly_pico['filepath'].values),
+    'stimulus_source': ('presentation', assembly_pico['identifier'].values),
+    'image_file_name': ('presentation', assembly_pico['image_file_name'].values),
+    'image_current_local_file_path': ('presentation', assembly_pico['image_current_local_file_path'].values),
+    'image_id': ('presentation', assembly_pico['stimulus_id'].values),
+    'repetition': ('presentation', assembly_pico['repetition'].values),
+    'stimulus_id': ('presentation', assembly_pico['stimulus_id'].values),
+    'filename': ('presentation', [assembly_pico['stimulus_id'].values[i] + '.png' for i in range(len(assembly_pico['stimulus_id'].values))]),
+    'background_id' : ('presentation',
+                    np.array([merged_stimulus_set.background_id[merged_stimulus_set.stimulus_id == stim_id].values[0] 
+                              for stim_id in assembly_pico['stimulus_id'].values])),
+    # selection of neuroid
+    'col' : ('neuroid', assembly_pico['col'].values),
+    'row' : ('neuroid', assembly_pico['row'].values),
+    'bank' : ('neuroid', assembly_pico['bank'].values),
+    'elec' : ('neuroid', assembly_pico['elec'].values),
+    'label' : ('neuroid', assembly_pico['label'].values),
+    'arr' : ('neuroid', assembly_pico['arr'].values),
+    'hemisphere' : ('neuroid', assembly_pico['hemisphere'].values),
+    'subregion' : ('neuroid', assembly_pico['subregion'].values),
+    'region' : ('neuroid', assembly_pico['region'].values),
+    'animal' : ('neuroid', assembly_pico['animal'].values),
+    'neuroid_id' : ('neuroid', assembly_pico['neuroid_id'].values),
+    # entire time_bin
+    'time_bin' : assembly_pico.time_bin
+}
+assembly_pico = xr.DataArray(data, dims=['presentation', 'neuroid', 'time_bin'],
+                        coords = coords)
+
+assembly_pico = NeuroidAssembly(assembly_pico)
+
+# some other operations for compatibility: we pad nans in Pico presentation coordinate to make sure that the number of repetitions is the same.:
+new_data = np.empty((197694, 75, 7))
+new_data[:] = np.nan
+
+new_assembly_pico = xr.DataArray(new_data, dims=['presentation', 'neuroid', 'time_bin'],
+                        coords = {
+                            'presentation': assembly_oleo.presentation,
+                            'neuroid' : assembly_pico.neuroid,
+                            'time_bin' : assembly_pico.time_bin
+                            })
+
+for i in tqdm.tqdm(range(len(assembly_pico)), desc='merging'): # 
+
+    r = i//35*28 +i
+    new_assembly_pico[r] = assembly_pico[i]
+
+
+# now we can finally merge the data:
+merged_data = np.concatenate((assembly_oleo.values, new_assembly_pico.values), axis=1)
+
+coords = {
+    # entire presentation
+    'presentation' : new_assembly_pico.presentation,
+    # selection of neuroid
+    'col' : ('neuroid', list(assembly_oleo['col'].values) + list(assembly_pico['col'].values)),
+    'row' : ('neuroid', list(assembly_oleo['row'].values) + list(assembly_pico['row'].values)),
+    'bank' : ('neuroid', list(assembly_oleo['bank'].values) + list(assembly_pico['bank'].values)),
+    'elec' : ('neuroid', list(assembly_oleo['elec'].values)+ list(assembly_pico['elec'].values)),
+    'label' : ('neuroid', list(assembly_oleo['label'].values) + list(assembly_pico['label'].values)),
+    'arr' : ('neuroid', list(assembly_oleo['arr'].values) + list(assembly_pico['arr'].values)),
+    'hemisphere' : ('neuroid', list(assembly_oleo['hemisphere'].values) + list(assembly_pico['hemisphere'].values)),
+    'subregion' : ('neuroid', list(assembly_oleo['subregion'].values) + list(assembly_pico['subregion'].values)),
+    'region' : ('neuroid', list(assembly_oleo['region'].values) + list(assembly_pico['region'].values)),
+    'animal' : ('neuroid', list(assembly_oleo['animal'].values) + list(assembly_pico['animal'].values)),
+    'electrode_id' : ('neuroid', list(assembly_oleo['neuroid_id'].values )+ list(assembly_pico['neuroid_id'].values)),
+    'neuroid_id' : ('neuroid', list(assembly_oleo['animal'].values + '_' + assembly_oleo['neuroid_id'].values) + list(assembly_pico['animal'].values + '_' + assembly_pico['neuroid_id'].values)),
+    
+    # entire time_bin
+    'time_bin' : assembly_oleo.time_bin
+}
+
+merged_assembly = xr.DataArray(merged_data,dims=['presentation', 'neuroid', 'time_bin'],
+                            coords=coords)
+
+merged_assembly = NeuroidAssembly(merged_assembly)
+merged_assembly = merged_assembly.transpose('presentation', 'neuroid', 'time_bin')
+
+merged_assembly.attrs = assembly_oleo.attrs
+
+merged_assembly.attrs['identifier']='Igustibagus2024'
+merged_assembly.attrs['stimulus_set']=merged_stimulus_set
+
+merged_assembly.name = 'Igustibagus2024'
+
+# package stimuli
+stimuli = StimulusSet(merged_stimulus_set)
+stimuli.stimulus_paths = {row['stimulus_id']: imgs_dir_path + '/' +  row['filename'] for _, row in stimuli.iterrows()}
+
+stimuli.drop('filename', axis=1, inplace=True)
+
+stimuli.name = 'Igustibagus2024'  # give the StimulusSet an identifier name
+
+assert len(stimuli) == 3138  # make sure the StimulusSet is what you would expect
+
+import pdb; pdb.set_trace()
+
+package_stimulus_set(catalog_name='brainio_brainscore', proto_stimulus_set=stimuli, stimulus_set_identifier=stimuli.name,
+                     bucket_name="brainio-brainscore") # upload to S3
+
+import pdb; pdb.set_trace()
+
+
+# upload assmebly to S3                      
+package_data_assembly('brainio_brainscore', merged_assembly, assembly_identifier=merged_assembly.name,
+                      stimulus_set_identifier=stimuli.name,assembly_class_name="NeuronRecordingAssembly", 
+                      bucket_name="brainio-brainscore")

brainscore_vision/data/igustibagus2024/merged_assembly/create_stimulus_set_with_background-id.py

@@ -0,0 +1,45 @@
+# import libraries 
+## general libraries 
+import pandas as pd
+
+## brain-score libraries
+import brainscore
+
+## domain-transfer specific libraries
+from helpers_background_id import loading_brain_model_activation
+
+
+##### NOTES ON THIS SCRIPT #####
+# This script is used to create a stimulus set with the background-id included. 
+# A fair question is: why do we need a model to do this? Shouldn't it depend only on the stimulus set?
+# the answer is: yes, it should. But this implementation takes the already existing from previous work 
+# of Marliawaty I Gusti Bagus, where the background is created as the presentation dimension of the 
+# brain activations extracted using the 'look_at' function from the brain-score package that is called 
+# on a stimulus set that doesn't have the background-id included.
+# It is however independent of the model used to extract brain activations.
+# FUTURE WORK should only use the stimulus set to create the background-id, and not the model activations.
+#################################
+
+# get the background-id
+brain_model_name='dcgan'
+image_source='dicarlo.domain_transfer'
+
+domain_transfer_data = loading_brain_model_activation(brain_model_name, image_source, penultimate_layer=False)
+
+background_df = pd.DataFrame({'id': domain_transfer_data.stimulus_id.values, 'background': domain_transfer_data.background_id.values})
+
+# include the background-id in the stimulus set
+stimulus_set = brainscore.get_stimulus_set('dicarlo.domain_transfer')
+
+dict_background = {}
+for row in range(len(background_df)):
+    dict_background[background_df.id[row]] = background_df.background[row]
+
+stimulus_set['background_id'] = [0] * len(stimulus_set)
+
+for i in range(len(stimulus_set)):
+    if stimulus_set.stimulus_id[i] in dict_background:
+        stimulus_set['background_id'][i] = int(dict_background[stimulus_set.stimulus_id[i]])
+
+
+stimulus_set.to_csv('merged_stimulus_set.csv', index=False)