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
3.35k stars 938 forks source link

Less data #21

Open sraddhanjali opened 6 years ago

sraddhanjali commented 6 years ago

I used your lstm blackbox for a small sized training dataset, and got very poor results? Could you suggest any ideas?

guillaume-chevalier commented 6 years ago

How much data is "less data" in your case?

It can depend on all sorts of things. For example, how long is your data, and how do you window it? How many features per time step, and how many classes? Are the classes imbalanced in quantity one compared to another?

sraddhanjali commented 6 years ago

The data is 3D accelerometer, gyroscope and linear acceleration data and classes are 23. All the time series have been interpolated to equal lengths, but no windowing or features are there per time step. The raw timeseries are used. There are 320 instances of each class for training and the testing has 10 instances of each class.