TemporalModelCommitments

.gitignore

@@ -0,0 +1,126 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+.scratch/
+.pytest_cache/
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+.idea/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don’t work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/

model_tools/brain_transformation/temporal_map.py

@@ -0,0 +1,127 @@
+from typing import Optional
+
+from brainio_base.assemblies import merge_data_arrays
+
+from model_tools.brain_transformation import LayerSelection
+
+from brainscore.model_interface import BrainModel
+from brainscore.metrics.regression import pls_regression
+
+from result_caching import store, store_dict
+
+class TemporalModelCommitment(BrainModel):
+    def __init__(self, identifier, base_model, layers, region_layer_map: Optional[dict] = None):
+        self.layers = layers
+        self.identifier = identifier or None
+        self.base_model = base_model
+        self.region_layer_map = region_layer_map or {}
+        self.recorded_regions = []
+
+        self.time_bins = None
+        self._temporal_maps = {}
+        self._layer_regions = None
+
+    def make_temporal(self, assembly):
+        assert self.region_layer_map																	# force commit_region to come before
+        assert len(set(assembly.time_bin.values)) > 1													# force temporal recordings/assembly
+
+        temporal_mapped_regions = set(assembly['region'].values)
+        assert bool \
+            (set(self.region_layer_map.keys()).intersection(temporal_mapped_regions))			        # force simalr brain regions
+
+        temporal_mapped_regions = list(set(self.region_layer_map.keys()).intersection(self.region_layer_map.keys()))
+        layer_regions = {self.region_layer_map[region]: region for region in temporal_mapped_regions}
+
+        stimulus_set = assembly.stimulus_set[assembly.stimulus_set['image_id'].isin(assembly['image_id'].values)]
+
+        stimuli_identifier = None
+        if self.identifier is not None:
+            stimuli_identifier = self.identifier + 'temporal_map_stim'
+
+        activations = self.base_model(stimulus_set, layers=list(layer_regions.keys()), stimuli_identifier=stimuli_identifier)
+        activations = self._set_region_coords(activations, layer_regions)
+
+        self._temporal_maps = self._set_temporal_maps(self.identifier, temporal_mapped_regions, activations, assembly)
+
+    def look_at(self, stimuli):
+        layer_regions = {self.region_layer_map[region]: region for region in self.recorded_regions}
+        assert len(layer_regions) == len(self.recorded_regions), f"duplicate layers for {self.recorded_regions}"
+        activations = self.base_model(stimuli, layers=list(layer_regions.keys()))
+
+        activations = self._set_region_coords(activations ,layer_regions)
+        return self._temporal_activations(self.identifier, activations)
+
+    @store(identifier_ignore=['assembly'])
+    def _temporal_activations(self, identifier, assembly):
+        temporal_assembly = []
+        for region in self.recorded_regions:
+            temporal_regressors = self._temporal_maps[region]
+            region_activations = assembly.sel(region=region)
+            for time_bin in self.time_bins:
+                regressor = temporal_regressors[time_bin]
+                regressed_act = regressor.predict(region_activations)
+                regressed_act = self._package_temporal(time_bin, region, regressed_act)
+                temporal_assembly.append(regressed_act)
+        temporal_assembly = merge_data_arrays(temporal_assembly)
+        return temporal_assembly
+
+    @store_dict(dict_key='temporal_mapped_regions', identifier_ignore=['temporal_mapped_regions', 'activations' ,'assembly'])
+    def _set_temporal_maps(self, identifier, temporal_mapped_regions, activations, assembly):
+        temporal_maps = {}
+        for region in temporal_mapped_regions:
+            time_bin_regressor = {}
+            region_activations = activations.sel(region=region)
+            for time_bin in assembly.time_bin.values:
+                target_assembly = assembly.sel(time_bin=time_bin ,region=region)
+                regressor = pls_regression(neuroid_coord=('neuroid_id' ,'layer' ,'region'))
+                regressor.fit(region_activations, target_assembly)
+                time_bin_regressor[time_bin] = regressor
+            temporal_maps[region] = time_bin_regressor
+        return temporal_maps
+
+    def _set_region_coords(self, activations, layer_regions):
+        coords = { 'region' : ( ('neuroid'), [layer_regions[layer] for layer in activations['layer'].values]) }
+        activations = activations.assign_coords(**coords)
+        activations = activations.set_index({'neuroid' :'region'}, append=True)
+        return activations
+
+    def _package_temporal(self, time_bin, region, assembly):
+        assert len(time_bin) == 2
+        assembly = assembly.expand_dims('time_bin', axis=-1)
+        coords = {
+            'time_bin_start': (('time_bin'), [time_bin[0]])
+            , 'time_bin_end': (('time_bin'), [time_bin[1]])
+            , 'region' : ( ('neuroid'), [region] * assembly.shape[1])
+        }
+        assembly = assembly.assign_coords(**coords)
+        assembly = assembly.set_index(time_bin=['time_bin_start', 'time_bin_end'], neuroid='region', append=True)
+        return assembly
+
+    def start_temporal_recording(self, recording_target, time_bins):
+        assert self._temporal_maps
+        assert self.region_layer_map
+        assert recording_target in self._temporal_maps.keys()
+        if time_bins is not None:
+            assert set(self._temporal_maps[recording_target].keys()).issuperset(set(time_bins))
+        else:
+            time_bins = self._temporal_maps[recording_target].keys()
+
+        self.recorded_regions = [recording_target]
+        self.time_bins = time_bins
+
+    def start_recording(self, recording_target):
+        assert self._temporal_maps
+        assert self.region_layer_map
+        assert recording_target in self._temporal_maps.keys()
+        if self.time_bins is None:
+            self.time_bins = self._temporal_maps[recording_target].keys()
+        self.recorded_regions = [recording_target]
+
+    def commit_region(self, region, assembly):
+        layer_selection = LayerSelection(model_identifier=self.identifier,
+                                         activations_model=self.base_model, layers=self.layers)
+        best_layer = layer_selection(assembly)
+        self.region_layer_map[region] = best_layer
+
+    def receptive_fields(self, record=True):
+        pass

tests/brain_transformation/test_temporal_map.py

@@ -0,0 +1,90 @@
+import pytest
+import functools
+import numpy as np
+from pytest import approx
+
+from model_tools.activations import PytorchWrapper
+from model_tools.brain_transformation.temporal_map import TemporalModelCommitment
+
+from brainscore import get_assembly
+from brainscore.benchmarks.loaders import AssemblyLoader, DicarloMajaj2015Loader, DicarloMajaj2015ITLoader
+
+# create a test assembly:
+class DicarloMajaj2015TemporalLoader(AssemblyLoader):                          # needed to add variation argument
+	def __init__(self, name='dicarlo.Majaj2015.temporal'):
+		super(DicarloMajaj2015TemporalLoader, self).__init__(name=name)
+		self._helper = DicarloMajaj2015Loader()
+
+	def __call__(self, average_repetition=True, variation=6):
+		assembly = get_assembly(name='dicarlo.Majaj2015.temporal')
+		assembly = self._helper._filter_erroneous_neuroids(assembly)
+		assembly = assembly.sel(variation=variation)
+		assembly = assembly.transpose('presentation', 'neuroid', 'time_bin')
+		if average_repetition:
+			assembly = self._helper.average_repetition(assembly)
+		return assembly
+
+def get_stim(assembly):
+	return assembly.stimulus_set[assembly.stimulus_set['image_id'].isin(assembly['image_id'].values)]
+
+def pytorch_custom():
+	import torch
+	from torch import nn
+	from model_tools.activations.pytorch import load_preprocess_images
+
+	class MyModel(nn.Module):
+		def __init__(self):
+			super(MyModel, self).__init__()
+			self.conv1 = torch.nn.Conv2d(in_channels=3, out_channels=2, kernel_size=3)
+			self.relu1 = torch.nn.ReLU()
+			linear_input_size = np.power((224 - 3 + 2 * 0) / 1 + 1, 2) * 2
+			self.linear = torch.nn.Linear(int(linear_input_size), 1000)
+			self.relu2 = torch.nn.ReLU()  # can't get named ReLU output otherwise
+
+		def forward(self, x):
+			x = self.conv1(x)
+			x = self.relu1(x)
+			x = x.view(x.size(0), -1)
+			x = self.linear(x)
+			x = self.relu2(x)
+			return x
+
+	preprocessing = functools.partial(load_preprocess_images, image_size=224)
+	return PytorchWrapper(model=MyModel(), preprocessing=preprocessing)
+
+class TestTemporalModelCommitment:
+	test_data = [(pytorch_custom, ['linear', 'relu2'], 'test_temporal', 'relu2', 'IT', 1, ['IT'], 39, 29)]
+	@pytest.mark.parametrize("model_ctr, layers, identifier, expected_best_layer, region, expected_region_count"
+							", expected_recorded_regions, expected_time_bin_count, expected_recorded_time_bin_cnt"
+							, test_data)
+	def test(self, model_ctr, layers, identifier, expected_best_layer, region
+			, expected_region_count, expected_recorded_regions, expected_time_bin_count, expected_recorded_time_bin_cnt
+			 # , expected_time_bins, expected_recorded_time_bins):
+			 ):
+		train_test_assembly_loader = DicarloMajaj2015TemporalLoader()
+		commit_loader = DicarloMajaj2015ITLoader()
+
+		training_assembly = train_test_assembly_loader(variation=3)
+		commit_assembly = commit_loader(average_repetition=False)
+		validation_assembly = train_test_assembly_loader(variation=6)
+
+		extractor = pytorch_custom()
+
+		t_bins = [t for t in training_assembly.time_bin.values if t[0] >= 0]
+
+		temporal_model = TemporalModelCommitment(identifier, extractor, layers)
+		# commit region:
+		temporal_model.commit_region(region, commit_assembly)
+		assert temporal_model.region_layer_map[region] == expected_best_layer
+		# make temporal:
+		temporal_model.make_temporal(training_assembly)
+		assert len(temporal_model._temporal_maps.keys()) == expected_region_count
+		assert len(temporal_model._temporal_maps[region].keys()) == expected_time_bin_count
+		# start recording:
+		temporal_model.start_temporal_recording(region, t_bins)
+		assert temporal_model.recorded_regions == expected_recorded_regions
+		# look at:
+		stim = get_stim(commit_assembly)
+		temporal_activations = temporal_model.look_at(stim)
+		assert set(temporal_activations.region.values) == set(expected_recorded_regions)
+		assert len(set(temporal_activations.time_bin.values)) == expected_recorded_time_bin_cnt