Orbit | Uber's New Open Source Python Library for Time Series Modeling

Orbit is Uber’s new python package for time series modeling and inference using Bayesian sampling methods for model estimation.

Orbit provides a familiar and intuitive initialize-fit-predict interface for working with time series tasks, while utilizing probabilistic modeling under the hood.

As per Orbit’s documentation, initial release supports concrete implementation for the following models:

• Local Global Trend (LGT)
• Damped Local Trend (DLT)

Both models, which are variants of exponential smoothing, support seasonality and exogenous (time-independent) features.

The initial release also supports the following sampling methods for model estimation:

• Markov-Chain Monte Carlo (MCMC) as a full sampling method
• Maximum a Posteriori (MAP) as a point estimate method
• Variational Inference (VI) as a hybrid-sampling method on approximate distribution

Quick Start

Installation

pip install orbit-ml

Orbit requires PyStan as a system dependency. PyStan is licensed under GPLv3 , which is a free, copyleft license for software.

Data

iclaims_example is a dataset containing the weekly initial claims for US unemployment benefits against a few related google trend queries from Jan 2010 - June 2018.

Number of claims are obtained from Federal Reserve Bank of St. Louis while google queries are obtained through Google Trends API.

Quick Starter Code

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

# import dependencies
import pandas as pd
import numpy as np
import warnings

from orbit.models.dlt import DLTMAP, DLTFull
from orbit.diagnostics.plot import plot_predicted_data

# Load data
df = pd.read_csv("../../examples/data/iclaims_example.csv", parse_dates=["week"])
df["claims"] = np.log(df["claims"])

# train test split
test_size = 104
train_df = df[:-test_size]
test_df = df[-test_size:]

# DLT model 
dlt = DLTMAP(
    response_col="claims",
    date_col="week",
    seasonality=52,
    seed=2020,
)

# fit
dlt.fit(df=train_df)

# predict
predicted_df = dlt.predict(df=test_df)

# plot
plot_predicted_data(training_actual_df=train_df, predicted_df=predicted_df,
                    date_col="week", actual_col="claims",
                    test_actual_df=test_df)

References