Finding the "normalize" function in MoSeq 2 code

[Lynnkoehler]

So I have been utilizing the transition matrices from my MoSeq outputs and I noticed that they all get normalized to 'bigram'.

I see in the code that they can also get normalized to 'row' or 'column' but I would like to see what sort of code/calculation is being used for this normalization so I can better understand the output.

For reference I am including a screen shot of the code where the normalize command is referenced.

[davidhbrann]

If you look a few lines down from the snippet you posted, you can see that normalize is passed to the get_group_trans_mats function from moseq2_viz, whose code you can find here. As you can see, the relevant details of the normalization can be further found within the moseq2_viz.model.get_transition_matrix and the moseq2_viz.model.normalize_transition_matrix functions. The normalization either divides by the total number of transitions (bigrams) or divides each row/column by their sums (to normalize with respect to incoming or outgoing transitions).

from moseq2_viz.model.trans_graph import get_trans_graph_groups, get_group_trans_mats

# syllable threshold defined above. Uncomment if you want to manually set
# max_syllables = 40

# select a transition matrix normalization method
normalize = 'bigram' # options: bigram, columns, rows

# load your model
model_path = progress_paths['model_path']
model_data = parse_model_results(model_path)
model_data['labels'] = relabel_by_usage(model_data['labels'], count='usage')[0]

# Get modeled session uuids to compute group-mean transition graph for each group
label_group, uuids = get_trans_graph_groups(model_data)
group = list(set(label_group))
# compute transition matrices and usages for each group
print('Group(s):', ', '.join(group))
trans_mats, usages = get_group_trans_mats(model_data['labels'], label_group, group, max_syllables, normalize=normalize)