How to get a predicted group list output? about machinelearning-samples HOT 9 CLOSED

@glebuk @Zruty0 can you help answer this query.

from machinelearning-samples.

@therollingbird,
The high-level approach to your problem can be solved via clustering as you mentioned. Please take a look at the iris clustering example here.
Basically, you first train a k-means clusterer with k=(number of groups you expect to end up) which is 5 in your example. Then, get the classId for each employee. Once each employee has one or more cluster ids assigned to them, just return them as a list, sorted by cluster.
Now, be aware, that this solution will give you groups of most similar people bucketized together. If you are trying to optimize teams' productivity, you might want to try do the exact opposite to ensure that there is a diversity of skills and proper creative tension.
Regards,
Gleb

from machinelearning-samples.

How would I do diversity with clustering ? That's one of the things I was trying to figure out. Either getting the groups and doing a manual input Oroing an inverse of the cluster algorithm if it's there

from machinelearning-samples.

@glebuk another way I'm thinking is using the knapsack algorithm I'm not sure if ML.net has it, if it does I think that would be the best solution.

If so my next question is does ML have the knapsack algorithm?

from machinelearning-samples.

What are you trying to optimize for each group? If similarity, then clustering is the correct approach.
Knapsack is not really machine learning problem. It is not surprising that ML.NET does not have it. It is an NP-complete optimization problem. There are a few pseudo polynomial approximation implementations using dynamic programming that can be found online.
Now, if you are trying to optimize diversity within each group, then simply do the reverse. Change the k to the size of each group (k=10 in your example) and form groups by taking a single member from each cluster per group.
In real life however, you will have to take into account psychological compatibility between members. How do you plan to compute that?

from machinelearning-samples.

Let me know if there are any further questions. Feel free to close the issue if it was answered.

from machinelearning-samples.

Hi that got me thinking thank you also one more question is there a way to select a number of elements I want in a cluster not just how many clusters I want? So instead of just selecting (k) clusters is there a way to select (n) elements?

from machinelearning-samples.

In general, training done over streaming data. We don't want to scan the entire data to count examples before training. I suggest you implement that logic in your code, by counting the number of training examples and then computing proper K. also note that we do not know apriori how many items will be in each cluster. As a result, that number is not known until training is done.

from machinelearning-samples.

Closing the issue/question as it is stale now.

from machinelearning-samples.