Python Tutorial: Machine learning and time series data

Want to learn more? Take the full course at https://learn.datacamp.com/courses/machine-learning-for-time-series-data-in-python at your own pace. More than a video, you'll learn hands-on coding & quickly apply skills to your daily work. --- In the final lesson of this chapter, we'll discuss the interaction between machine learning and timeseries data, and introduce why they're worth thinking about in tandem. First, let's give a quick overview of the data we'll be using. They're both freely available online, and come from the excellent website Kaggle-dot-com. Audio is a very common kind of timeseries data. Audio tends to have a very high sampling frequency (often above 20,000 samples per second!). Our first dataset is audio data recorded from the hearts of medical patients. A subset of these patients have heart abnormalities. Can we use only this heartbeat data to detect which subjects have abnormalities? Audio data is often stored in "wav" files. We can list all of these files using the "glob" function. It lists files that match a given pattern. Each of these files contains the auditory data for one heartbeat session, as well as the sampling rate for that data. We'll use a library called "librosa" to read in the audio dataset. Librosa has functions for extracting features, visualizations, and analysis for auditory data. We can import the data using the "load" function. The data is stored in audio and the sampling frequency is stored in sfreq. Note that the sampling frequency here is 2205, which means 2205 samples are recorded per second. Using only the sampling frequency, we can infer the timepoint of each datapoint in our audio file, relative to the start of the file. Now we'll create an array of timestamps for our data. To do so, you have two options. The first is to generate a range of indices from zero to the number of datapoints in your audio file, divide each index by the sampling frequency, and you have a timepoint for each data point. The second opt