Next Episode: Coming Soon!

Date: TBD
Time: - TBD

Description: Check back soon for more details

SHAP Deep Dive

Tonight we're going to conclude our final episode of this series by taking a deep dive into the SHAP source code. The previous two workshops discussed the 'what', and this final one will cover the 'why'. It will involve looking at the literature that gave birth to SHAP, as well as rebuilding our own tree explainer!

SHAP Continued: Understanding Interaction Effects on Policing

In today's workshop we'll finish up our discussion of SHAP and how we can use it to understand the inner workings of complicated Machine Learning models. Today's lesson will focus a little more on the "how" behind SHAP and we'll take a look at some of its source code. We'll also take a look at some of the more advanced plots it can generate and what they mean. We'll continue using the North Carolina policing dataset as our study tool.

Using SHAP to Deconstruct Policing Outcomes

In today's workshop we'll continue our exploration of policing outcomes by introducing a new and relevant breakthrough in interpetable ML: SHAP values! SHAP is a technique you can use to breakdown the prediction of any ML model into its contributing parts. It's a great way to make black box models interpretable to a wide audience. For today, we'll take our pre-built model and use SHAP to figure out the distribution of importance of different features, and build very handy waterfall charts to deconstruct individual predictions.

Modeling Challenge: Predict Policing Outcomes II

In today's webinar we'll do another rehash of gradient boosting, and learn how it will be applied to a large dataset of over 4.5 million police stops!

Modeling Challenge: Predict Policing Outcomes

In today's workshop we're going to work with a dataset of 4.5 million police stops to study the impact of different characteristics on police outcomes. Using data from the open policing project, we'll study how police outcomes are influenced by characteristics such as someone's race, gender, and behavioral characteristics of the stop itself. It's designed to give attendees working practice using different modeling techniques on an interesting dataset of real world importance.

Dashboarding -- Final Deployment

In our final session for this series, we'll take our covid dashboard and finally deploy it to the web. With our automated model fitting setup, we'll go ahead and take our dashboard and put it on the web for all of the world to see! This will give everyone who followed along a unique and beneficial capstone to demonstrate their time commitment throughout the project.

Dashboarding -- Automated Modeling

In today's workshop we'll start go beyond building chart with callbacks, and turn our attention to populating it with statistical models! By incorporating statistical analysis into our final product, this will truly be a data science product, and not merely an enhanced BI tool. We'll work through sample scripts that connect to data and create predictions, and walk through the critical steps of running them on an automated schedule that will put model results into a database for future use.

Dashboarding -- Designing Your Dashboard

In today's workshop we're going to get a deep dive on how we can make our dashboards look presentable. The class will be spent going over css grids available in Dash, and how to use them to design layouts.

Dashboarding With Plotly -- Callbacks II

In our continuing series on dashboarding, we'll enhance our covid dashboard forecasting app by integrating more callbacks to make it more interactive. We'll also begin looking at how to control and modify layout with css grids.

Dashboarding With Plotly -- Callbacks

In today's workshop, we'll continue our discussion of dashboarding, and take a deeper dive into the main tool used to add interactivity to dashboards: callbacks. We'll add more charts to our dashboards, and add three additional user inputs that are chained together to create an interesting and interactive charting experience that allows users to see how covid metrics have changed across time and space.

Dashboarding -- Dash Intro

In today's webinar we'll build off of our previous lesson and take our plot.ly plots and turn them into our first dashboard! The main thrust of today's class will be to understand the basics of Dash's callback system, and how we can use them to trigger changes to particular parts of a web page when user actions are taken. By the end of the class we'll have a simple but functional dashboard that allows us to interactively look at covid metrics across the US.

Dashboarding -- Plotly Intro

Welcome to our newest webinar series: dashboarding! Dashboarding is probably the single most useful tool for an aspiring data science practitioner. All projects eventually need to be turned into a usable dashboard in order for their impact to be fully realized. We'll begin the current webinar with an overview of dash & plotly, and will have a long form lab that allows us to use the plotly visualization library, which is required for visualizations within dash.

Forecasting With Prophet -- Bayesian Deep Dive II

In our final wrap up of our series on forecasting, we'll continue with our discussion of STAN, the bayesian library used to generate the data for Prophet models. With the previous webinar behind us, we'll look to take a more applied approach and use MCMC techniques to generate our own custom models, including the ones that Prophet uses to make its predictions. Students should expect to have an introductory outline of how bayesian modeling works, and what you might be able to use it for.

Forecasting With Prophet -- Bayesian Deep Dive

In today's class we'll take our time with Forecasting in a different direction: looking at how Prophet generates its predictions in the first place. Prophet is actually a wrapper around a more powerful, low-level library called STAN, which is used for generating probabilistic data for bayesian models. For this class we're going to take a step back from our applied approach, and use the last sections of this segment to gain a better understanding of how bayesian programming works. Today's webinar will be a crash course in bayesian concepts, and provide some groundwork for being able to generate our own probabilistic models.

Forecasting With Prophet -- Forecasting Product Sales Pt 2

In our final applied webinar for this module, we'll take a second look at our dataset of product sales, and this time project our predictions across several levels of hierarchy at the same time. We'll compare how our predictions for our stores compare to their individual products, and explore possibly improved approaches.

Hierarchical Time Series -- Product Sales

In today's webinar we continue our discussion of Hierarchical Time Series -- but this time we will change direction and start using a different dataset. With the last 3 classes behind us, we'll see how we've retained the lessons by working on an extended lab where we predict product sales from top level sales numbers that are then progressively grouped down into lower levels. This week's class will be more of an extended lab that will not focus so much on new material, but rather give everyone a longer form, extended lab to work through previous problems and solidify concepts in their own mind.

Forecasting With Prophet -- Hierarchical Time Series III

In this workshop we'll continue our discussion of hierarchical time series -- when you have a forecast that needs to be made on multiple levels. We'll continue with our dataset of covid19 cases. Up until this point we've used the 'top down' method, where we project a high level forecast into lower parts. Today we'll do the exact opposite -- fit time series models for each of the individual states and aggregate them into a larger forecast for the country as a whole. Image provided courtesy of <a href="http://www.nyphotographic.com/">Nick Youngson</a>

Hierarchical Time Series with Prophet -- Continued

In today's webinar, we continue where we left off from the previous class -- building out a hierarchical time series model from the 'top down'. Ie, starting with high-level forecast from a single time series and breaking it down into smaller subcomponents. The main thrust of the lesson goes over how to project our highest level forecast onto the individual states. In doing so, we will test several different strategies for allocating country-wide covid data between the different states.

Forecasting With Prophet - Hierarchical Time Series

In today's webinar, we'll take our discussion of Prophet and forecasting one step further: how do we build time series models when the data is broken into many subcomponents? Ie, what if we wanted to predict overall sales from individual regions? Attendance from individual locations, and so on. To continue with our usage of covid19 data, we'll now turn our attention to building predictions from individual states to come up with our aggregate predictions.

Forecasting With Prophet - Intermediate Functionality

In this follow up webinar to our previous discussion of Prophet, we'll turn our attention to more nuanced problems people frequently have to deal with when working on forecasting: breakpoints in your data, non-time based variables, and adding custom holidays & seasonality. We will continue working with covid19 data, and explore whether or not accounting for the George Floyd protests improves our results, and how adding data about hospitalizations and infections allows for better forecasts.

Forecasting With Prophet!

Prophet is a powerful, bayesian based forecasting tool for quickly developing accurate and interpretable forecasts. Forecasting is a common business use case for ML and Data Science practitioners but is often undercovered within the industry. Being able to quickly put together accurate and useful forecasts is something most businesses are unable to do well. This forecasts will introduce students to the basics of the prophet library, how to use it to build forecasts and interpret your results. We'll take covid-19 infection and case data, and use our lab to build predictive models for future events!