https://github.com/wd60622/latent-calendar/actions/runs/7440016777/job/20240552324?pr=31#step:5:142

Either some cadence or only a subset of the data

The docs here are not rendering properly

https://wd60622.github.io/latent-calendar/modules/plot/elements/#latent_calendar.plot.elements.CalendarEvent

Using sns.heatmap to plot each calendar would introduce all the features from the heatmap. This would include:

• annotated values
• automatic cmap

https://seaborn.pydata.org/generated/seaborn.heatmap.html#seaborn.heatmap

Storing columns of wide format limits generalization to more time slots per day

MulitIndex would allow for pandas support of

• summing across
• pivoting / reshaping

Need to ensure compat with sklearn and the DOWHour / segments

Requires a bit more work than you'd think but might be worth the effort

Will have to direct into docs folder

Loosen this up by making it as an optional install part of the package.

It is only used in

• data generation at the moment

The stack_segments takes a list of Series but might benefit from list of DataFrames or Series

mondays = ...
tuesday = ... 

df_segments = stack_segments([mondays, tuesdays])

weekends = ... 
df_segments_with_weekends = stack_segments([df_segments, weekends])

More details on the problem formulation and how it relates to other problems

Error when using with python 3.11

Potential fix:

• have TimeLabeller use default_factory
• migrate away from dataclasses?

>>> import latent_calendar
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.11/site-packages/latent_calendar/__init__.py", line 1, in <module>
    import latent_calendar.extensions
  File "/usr/local/lib/python3.11/site-packages/latent_calendar/extensions.py", line 85, in <module>
    from latent_calendar.plot.core import (
  File "/usr/local/lib/python3.11/site-packages/latent_calendar/plot/__init__.py", line 14, in <module>
    from latent_calendar.plot.core import (
  File "/usr/local/lib/python3.11/site-packages/latent_calendar/plot/core/__init__.py", line 1, in <module>
    from latent_calendar.plot.core.calendar import (
  File "/usr/local/lib/python3.11/site-packages/latent_calendar/plot/core/calendar.py", line 14, in <module>
    from latent_calendar.plot.elements import (
  File "/usr/local/lib/python3.11/site-packages/latent_calendar/plot/elements.py", line 31, in <module>
    @dataclass
     ^^^^^^^^^
  File "/usr/local/lib/python3.11/dataclasses.py", line 1230, in dataclass
    return wrap(cls)
           ^^^^^^^^^
  File "/usr/local/lib/python3.11/dataclasses.py", line 1220, in wrap
    return _process_class(cls, init, repr, eq, order, unsafe_hash,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dataclasses.py", line 958, in _process_class
    cls_fields.append(_get_field(cls, name, type, kw_only))
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dataclasses.py", line 815, in _get_field
    raise ValueError(f'mutable default {type(f.default)} for field '
ValueError: mutable default <class 'latent_calendar.vocab.HourFormatter'> for field hour_formatter is not allowed: use default_factory

How the LatentCalendar class can be used to integrate with sklearn via Pipeline or ColumnTransformer

Can components be initialized using sklearn?

The CalendarEvent does not support rectangles across multiple days

This might be single block from Monday until Friday from 9 - 5

https://www.nyc.gov/site/tlc/about/tlc-trip-record-data.page

Location IDs
https://data.cityofnewyork.us/Transportation/NYC-Taxi-Zones/d3c5-ddgc

Normalize takes a string argument, separate into different methods instead

Would be nice to work with conditional probabilities:

P[H=h | D=d, N=n]
P[D=d | H=h, N=n]

Derived from P[H=h, D=d | N=n]

• GitHub action for testing different python versions

df = pd.DataFrame({
    "hour": np.linspace(0, 24, 100)[:-1]
})

# Like before (rounded to nearest hour)
# 0.75 -> 0, 23.75 -> 23
discretizer = HourDiscretizer(col="hour")
discretizer.transform(df)

# 0.75 -> 0.5, 23.75 -> 23.5
discretizer = HourDiscretizer(col="hour", minutes=30)
discretizer.transform(df)

Affected here and would want to be available where this functionality is exposed.

I initially thought it was at Invoice level rather than transaction level

https://github.com/wd60622/latent-calendar/actions/runs/7440016777/job/20240552586?pr=31#step:5:154

Create some tests for the plotting outcomes with pytest-mpl

Need a fully Bayesian LDA in order to make use of custom priors.

Building off the examples here

# (n_components, n_time_slots)
prior = df_segments.to_numpy()

from latent_calendar import BayesianLatentCalendar

model = BayesianLatentCalendar(prior=prior)
model.fit(df_model)

The BaseEstimator has a mixin to link to the docs
https://github.com/scikit-learn/scikit-learn/blob/8721245511de2f225ff5f9aa5f5fadce663cd4a3/sklearn/utils/_estimator_html_repr.py#L406

This can be used to link to the latent calendar models
https://wd60622.github.io/latent-calendar/modules/model/#latent_calendar.model.latent_calendar.LatentCalendar
https://wd60622.github.io/latent-calendar/modules/model/#latent_calendar.model.latent_calendar.ConjugateModel
https://wd60622.github.io/latent-calendar/modules/model/#latent_calendar.model.latent_calendar.MarginalModel
https://wd60622.github.io/latent-calendar/modules/model/#latent_calendar.model.latent_calendar.DummyModel

Older versions of matplotlib do not support the rotation argument

Happens here

changes in matplotlib>=3.5.0 which is about 3 years old

Badges for tests, docs, and pypi version

The way that scikit learn does it : https://github.com/scikit-learn/scikit-learn/blob/8721245511de2f225ff5f9aa5f5fadce663cd4a3/sklearn/base.py#L478-L497

Might be something to do with the version as the new tests in #64 pass with fresh environment

Is your feature request related to a problem? Please describe.

anything that is fit should have the _ after the variable

Based on LDA, the literature suggest value <= 1 which is part of the current sklearn implementation. It might be good to relax for this use case