CODEWITHSUNDEEP
pythononline-machine-learningcreme
Leandro Filipe
Leandro Filipe

Reputation: 1

Blind Position Online Prediction

new here. I'm currently working on my MSc thesis that consists of developing a set of blinds/shutters that predict the user's behavior and set the position automatically, according to the user's habits and likes. I'm using Online Learning to solve my problem, specifically the library creme (python). The problem is I can't get good enough results, with an MAE of around 24, which is way too much. Do you guys have any experience/ideas on how to solve a problem like this? I'm getting these numbers with a model like this:

model = compose.Discard('Year', 'Timestamp')
model += feature_extraction.Agg(on='Radiation', by='Month', how=stats.Mean())
model |= linear_model.PARegressor()
model = time_series.Detrender(regressor=model, window_size=105120) #Around a year

And with the following features:

Being Position the label

Upvotes: 0

Views: 56

Answers (1)

Max Halford
Max Halford

Reputation: 11

I'm the creator of creme. Your task looks very interesting!

I recommend that you open a dedicated issue on GitHub. In particular, we will need access to a dataset in order to reproduce the performance you're describing.

Upvotes: 0

Related Questions