Help with homework

[Meowmeow]

Just now, ring_master said:

K- means vaadu but K-means assigns cluster to it's members based on euclidean distance between clusters

tSNE is what I used for clustering in the earlier problem based on Euclidian distance. 

[kathanayaka]

First use gridsearch and randomsearchcv algorithms to get the best hyper parameters

Then try K-means(n=1,2,3,....) and SVM(best model) for the clustering(playing with c,gamma values)

Hierarchial clustering before k means will also give you an estimate on optimal no of clusters

[Meowmeow]

Just now, kathanayaka said:

First use gridsearch and randomsearchcv algorithms to get the best hyper parameters

Then try KNN(n=1,2,3,....) and SVM(best model) for the clustering(playing with c,gamma values)

Does that model cluster based on Euclidean distance between the clusters of the probability distribution of the cluster? 

[kathanayaka]

2 minutes ago, Meowmeow said:

Does that model cluster based on Euclidean distance between the clusters of the probability distribution of the cluster? 

K means supports only Euclidean distance
 

[ring_master]

3 minutes ago, Meowmeow said:

tSNE is what I used for clustering in the earlier problem based on Euclidian distance. 

Clustering will be based on similiarity measure... so distance is used as similarity measure . You don;t want to use distance to measure similarity? Is that you're trying to do?

[Meowmeow]

2 minutes ago, kathanayaka said:

K means supports only Euclidean distance
 

Yeah I used tSNE for Euclidian distance. I need an algorithm for probability distribution of the cluster. 

[kathanayaka]

Just now, Meowmeow said:

Yeah I used tSNE for Euclidian distance. I need an algorithm for probability distribution of the cluster. 

why did even tSNE come into picture

Its just for data exploration and visualizing high-dimensional data and gives an intuition of how the data is arranged ante

[Ellen]

Just now, Meowmeow said:

Yeah I used tSNE for Euclidian distance. I need an algorithm for probability distribution of the cluster. 

There are some divergence based clustering. Special cases which measure probability distribution. try kl-divergence 

or u  can simply use chi-squared 

[Meowmeow]

1 minute ago, ring_master said:

Clustering will be based on similiarity measure... so distance is used as similarity measure . You don;t want to use distance to measure similarity? Is that you're trying to do?

Kind of, if I cluster based on Euclidean distance, I can only tell that the points in a single cluster are similar. But I cannot tell how different one cluster is from another cluster based on how close/far it is. So, I am trying to get that information. 

[Meowmeow]

1 minute ago, kathanayaka said:

why did even tSNE come into picture

Its just for data exploration and visualizing high-dimensional data and gives an intuition of how the data is arranged ante

I am trying to visualize high dimensional data, but I want the distance between the clusters to indicate how similar/or not they are to each other. 

[Meowmeow]

I might just give up this question, this is for extra credit and I think I dided everything else right. 

[Scada]

Just now, Meowmeow said:

I might just give up this question, this is for extra credit and I think I dided everything else right. 

Final ga nuv bussu anna mata..

[kathanayaka]

5 minutes ago, Meowmeow said:

I am trying to visualize high dimensional data, but I want the distance between the clusters to indicate how similar/or not they are to each other. 

yes you need a clustering algorithm . hence K means or SVM give a shot

[ring_master]

8 minutes ago, Meowmeow said:

Kind of, if I cluster based on Euclidean distance, I can only tell that the points in a single cluster are similar. But I cannot tell how different one cluster is from another cluster based on how close/far it is. So, I am trying to get that information. 

Ok I got it . 

[Meowmeow]

3 minutes ago, kathanayaka said:

yes you need a clustering algorithm . hence K means or SVM give a shot

Arey k means clustering answer kadu ra, this is a traditional approach he thought in the class, I used this algorithm for all the earlier problems.