Time series analysis using Prophet in Python — Part 2: Hyperparameter Tuning and Cross Validation
In the previous article, we explained all the math behind Prophet. This article, I am going to show you how to use Prophet in practice and how to do hyperparameter tuning.
Dataset
Let’s start with a real time series dataset. Since we are using Prophet, let’s see how many times Prophet has downloaded from PyPI. Here we get the daily download counts by running this Google Bigquery command in subprocess (Note that you’ll need to install Google Cloud SDK https://cloud.google.com/sdk/docs/downloads-yum to be able to run the bq command to query data from Bigquery):
Let’s load the data. Here we can see that there are 637 observations. The date is from 2018–10–01 to 2020–06–30 as we have defined in the query. The max number of daily downloads is 32258. (We didn’t query any data before 2018–10–01 is because data before that is not reliable in PyPI).
