```custom-WatchMotionIndex
activePost: 2
```

Now to the fun part: Getting a lot of training data! I'm trying to collect at least a couple recordings of each exercise every day. My plan is to collect enough data to train a simple model that I can implement in the watch app and do some basic exercise classification on device just to see that it works.

When building a machine learning model to classify exercises, the quality of the dataset is just as important as the algorithm. A balanced dataset ensures that the model learns equally from all exercise types, rather than favoring one over the others. To measure how balanced the dataset is, we can use the coefficient of variation (CV) which simply is the standard deviation divided by the mean. We want the CV to be as low as possible, if the dataset is balanced, the CV will be close to 0.

```custom-WatchMotionMoreTrainingData

```

The data is updated every time I add a new recording to the dataset, so check back later to see how it progresses!


More training data

Watch Motion

```custom-WatchMotionIndex
activePost: 1
```

![{2400[x]1600}{priority}](/journal/241109/watch-motion-data-collection-2.jpg)

With the app working and the data collection endpoint ready I started collecting data. I made some small changes to the app so I can select which exercise I'm doing before starting the data collection.

There's also a 5-second delay before the recording starts to account for the time it takes to get started doing some reps. The samples are collected at 100Hz and by default for 10 seconds.

My first test from last weekend with situps looked good so I will keep that as one exercise. I also added overhead press, chins and battle ropes. The hardest one to predict will probably be chins as the wrists doesn't move that much during the exercise.

With that figured out, it was time to head out to the gym and start collecting some real world data!

<iframe
  className="video-1-2 mb-8"
  src="https://www.youtube.com/embed/IEwDTtB-taE?si=URbvqm8xEbkwJ8Rg"
  title="YouTube video player"
  allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
  allowFullScreen
></iframe>

Here are some of the samples I collected for the four exercises. This is unprocessed data straight from the watch, I will most likely need to do some processing and normalization before I can use it for training.

```custom-WatchMotionDataCollection

```


Data collection

```custom-WatchMotionIndex
activePost: 0
```

![{2400[x]1600}{priority}](/journal/241101/watch-motion-getting-started-1.jpg)

Started the weekend with a new project. It's an Apple Watch app that
collects and analyzes motion data from your daily activities. By leveraging
the watch's built-in sensors, it tracks your movements throughout the day to
provide insights about your physical activity patterns and movement quality.

The data is sent to a serverless function that stores it in a database and
exposed via a REST API to be visualized here. Made a small test in the
outdoor gym and it worked pretty well. Next step is probably to see what I
can do with the data, it would be pretty cool to be able to predict exercises
for example.

<iframe
  className="video-1-2 mb-8"
  src="https://www.youtube.com/embed/WjQSIbHgatA?si=RpozVyp4HzGUtM2F"
  title="YouTube video player"
  allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
  allowFullScreen
></iframe>

I collected the data for 20 seconds with a sampling frequency of 50Hz, and this is what it looks like:

```custom-WatchMotionGettingStarted

```


Watch Motion

Data collection

Project: Watch Motion