Latent Calendar

Analyze and model data on a weekly calendar

Installation

Install from PyPI:

pip install latent-calendar

Or install directly from GitHub for the latest functionality.

Features

• Integrated automatically into pandas with cal attribute on DataFrames and Series
• Compatible with scikit-learn pipelines and transformers
• Transform and visualize data on a weekly calendar
• Model weekly calendar data with a mixture of calendars
• Create lower dimensional representations of calendar data

Quick Start

from latent_calendar.datasets import load_chicago_bikes
from latent_calendar import LatentCalendar
from latent_calendar.plot import plot_model_components

import matplotlib.pyplot as plt

df_trips = load_chicago_bikes()
df_stations = df_trips.cal.aggregate_events("start_station_name", "started_at", minutes=60)

model = LatentCalendar(n_components=10, random_state=42)
model.fit(df_stations)

order = model.component_distribution_.argsort()[::-1]
plot_model_components(model, components=order)
first_date, last_date = df_trips["started_at"].min().date(), df_trips["started_at"].max().date()
title = f"Trip behavior between {first_date} and {last_date} (ordered by component frequency)"
plt.suptitle(title)