LSTM
Understand the structure and function of LSTM neural networks for sequential data analysis. Learn to preprocess time series data by reshaping and scaling, add lagged data inputs, build LSTM models in TensorFlow, and evaluate their forecasting performance. This lesson prepares you to apply deep learning techniques to time series forecasting.
Understanding LSTM
LSTM stands for long short-term memory and is a type of neural network mostly used for sequential data (such as time series and text data) in which the order of the inputs matter. Before we dive into LSTM, let's take a quick look at neural networks to understand how they work.
A neural network is a computer system modeled on the human brain and nervous system. It can have multiple layers, often divided into three categories—input, hidden, and output. Each of these layers is composed of several nodes that work as functions, receiving input and sending output between each other.
The difference between LSTM and a standard neural network is that it also has feedback connections, meaning that each node can also use its output from the past as input for the present, allowing it to process entire sequences of data.
From the image above, we can see that each LSTM cell uses its own past output as input (