Model inclusion of track graph in Decoding notebooks

[CBroz1]

This line mentions the possibility of including a track graph in decoding

spyglass/notebooks/py_scripts/41_Decoding_Clusterless.py

Lines 259 to 261 in 03e3996

	# ### 1D Decoding
	#
	# If you want to do 1D decoding, you will need to specify the `track_graph`, `edge_order`, and `edge_spacing` in the `environments` parameter. You can read more about these parameters in the [linearization notebook](./24_Linearization.ipynb). You can retrieve these parameters from the `TrackGraph` table if you have stored them there. These will then go into the `environments` parameter of the `ContFragClusterlessClassifier` model.

In my brief attempt, I did not find the process of including this graph straight forward, needing to remap kwargs across a fetched value and classes from non_local_detector. I hit errors with the embedded ObservationModel having an empty name

Partial error stack

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
src/spyglass/utils/dj_mixin.py:612: in populate
    return super().populate(*restrictions, **kwargs)
../datajoint-python/datajoint/autopopulate.py:254: in populate
    status = self._populate1(key, jobs, **populate_kwargs)
../datajoint-python/datajoint/autopopulate.py:322: in _populate1
    make(dict(key), **(make_kwargs or {}))
src/spyglass/decoding/v1/clusterless.py:212: in make
    classifier.fit(
../../miniconda3/envs/spy/lib/python3.9/site-packages/non_local_detector/models/base.py:960: in fit
    self._fit(
../../miniconda3/envs/spy/lib/python3.9/site-packages/non_local_detector/models/base.py:493: in _fit
    self.initialize_state_index()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = ClusterlessDetector(clusterless_algorithm='clusterless_kde',
                    clusterless_algorithm_params={'block_...ke=False)],
                    sampling_frequency=500.0,
                    state_names=['Continuous', 'Fragmented'])

    def initialize_state_index(self):
        self.n_discrete_states_ = len(self.state_names)
        bin_sizes = []
        state_ind = []
        is_track_interior = []
        for ind, obs in enumerate(self.observation_models):
            if obs.is_local or obs.is_no_spike:
                bin_sizes.append(1)
                state_ind.append(ind * np.ones((1,), dtype=int))
                is_track_interior.append(np.ones((1,), dtype=bool))
            else:
                environment = self.environments[
>                   self.environments.index(obs.environment_name)
                ]
E               ValueError: '' is not in list

../../miniconda3/envs/spy/lib/python3.9/site-packages/

[edeno]

Not quite sure what you're running here. It seems like the environment name given for the observation model is not one specified in the list of Environments given to the model.

[CBroz1]

Run snippet

@pytest.fixture
def some_name(track_graph):
    from non_local_detector.environment import Environment
    from non_local_detector.models import ContFragClusterlessClassifier
    from spyglass.decode import v1 as decode_v1

    graph_entry = track_graph.fetch1() # Restricted table
    class_kwargs = dict(
        clusterless_algorithm_params={
            "block_size": 10000,
            "position_std": 12.0,
            "waveform_std": 24.0,
        },
        environments=[
            Environment(
                environment_name=graph_entry["track_graph_name"],
                track_graph=track_graph.get_networkx_track_graph(),
                edge_order=graph_entry["linear_edge_order"],
                edge_spacing=graph_entry["linear_edge_spacing"],
            )
        ],
    )
    params_pk = {"decoding_param_name": "contfrag_clusterless"}
    decode_v1.core.DecodingParameters.insert_default()
    decode_v1.core.DecodingParameters.insert1(
        {
            **params_pk,
            "decoding_params": ContFragClusterlessClassifier(**class_kwargs),
            "decoding_kwargs": dict(),
        },
        skip_duplicates=True,
    )

[edeno]

If you omit: environment_name=graph_entry["track_graph_name"], then it should work, otherwise you have to specify the same name in the ObservationModel.