Trump Tweets (Fun with Deep Learning)
In this talk I will introduce LSTMs. This is how deep learning deals with time series data. The dataset we will be focusing on is Trump's tweets. Using this data we will make a tweet generator in which we will train how to simulate Trump style tweets one character at a time.
We will be using Keras to generate the Deep Learning model. Google colab will be used so that all attendees will be able to use a GPU.
Outline/Structure of the Talk
The initial session will focus on cleaning the data since trump has a tendency to use emojis. Hence some data cleansing will occur in pandas.
Much of the talk will however focus on building a time series model in Keras. I will show how a few lines is all that is required to build a Deep Learning model.
If there is time remaining we shall use a extremely similar model to classify fake news.
Learning Outcome
The goal is for the attendees to learn to deal with time series data in a Deep Learning context.
Target Audience
People who have some experience in basic statistics. Some experience in Deep Learning will help but is not necessary.
Links
https://github.com/sachinruk/deepschool.io (Lesson 16 and 18)