search

guillaume-chevalier / LSTM-Human-Activity-Recognition

Human Activity Recognition example using TensorFlow on smartphone sensors dataset and an LSTM RNN. Classifying the type of movement amongst six activity categories - Guillaume Chevalier
MIT License
3.33k stars 935 forks source link

How did you decide the number of hidden layers to be 32. Can you please explain this? #35

Open sidharthgurbani opened 4 years ago

sidharthgurbani commented 4 years ago

You have 128 readings in every window. How will these input values go into the LSTRM network? I am trying to follow similar logic for my own generated dataset. So I need to understand the logic to implement for my dataset.

guillaume-chevalier commented 3 years ago

How did you decide the number of hidden layers to be 32. Can you please explain this?

As of having tried ANNs and RNNs on other datasets before, I know that with this quantity of data, this amount of hidden layer would be good. In practice, I've tried a few ones before settling on one. You could use this to speed up this process of finding the best parameters: https://www.neuraxle.org/stable/hyperparameter_tuning.html

You have 128 readings in every window. How will these input values go into the LSTM network?

Using Perceptrons and a 3D cube. Values are normalized and inputted to the perceptrons of the LSTM. The axis of time is well separated as a specific dimension in the 3D cube.