Creating a clustering model using simple k-means algorithm

TensorFlow General Discussion
,
1

Firstly very new to Tensorflow and Keras
Installed via Anaconda Navigator - very straight forward

I wish to create the above described model using Tensorflow BUT I am having difficulty in understanding the following…

My input is a time series - 5 consecutive hourly prices held in an excel worksheet
first row is at time t
second row is at time t+1 etc.

I wish k to be 25 - i.e. 25 clusters

But I want the clustering to be based on the graph/plot of the 5 prices not the actual prices themselves as per the following example

example1

I have already tried this exercise in weka but the clustering resulted in 25 straight lines based on the actual values of the prices and NOT on the graph/plot

Could somebody please explain the difference between the two styles of clustering and shoe me how to achieve the clustering I want

Thank you

Bob M

3

Hello @Bob_Matthews

Thank you for using TensorFlow,
Using Auto Encoders and Kmeans algorithm we can make shape based clustering possible, So here encoder acts as a feature extractor and then for K=25 clustering algorithm is applied.
If we use simple preprocessing and Kmeans algorithm, the data should be normalized and also should be scaled before applying Kmeans algorithm.
Thank you