prediction of wind speed

[TUSHAR VAGHASIYA]

I have wind speed data of last one month in this format (30 days(Row) and 24 hours(column)). Now, I want to predict wind speed data of one-day(hour by hour) by using matrix factorization technique. I use a visual studio software in windows system.

Anybody could please give me the suggestion for it?

[Zeno Gantner]

Hi,

I am not sure whether MML is the right tool for this task.

How is this related to personalization?

Also, it seems your data matrix is sparse, right?

Cheers,

   Z.

On Mon, Mar 20, 2017 at 10:23 AM, TUSHAR VAGHASIYA <tkvagha...@gmail.com> wrote:

I have wind speed data of last one month in this format (30 days(Row) and 24 hours(column)). Now, I want to predict wind speed data of one-day(hour by hour) by using matrix factorization technique. I use a visual studio software in windows system.

Anybody could please give me the suggestion for it?

--
You received this message because you are subscribed to the Google Groups "MyMediaLite" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mymedialite+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Tushar Vaghasiya]

Hi,

Thank you very much for your message.

As we predict the rating of item by using matrix factorization.

can't we predict the wind speed by using matrix factorization?

I have attached the format of wind speed for one week. now i want to predict 8th day wind speed hour by hour.

could you please suggest me?

Thank you very much in advance.

Regards,

Tushar

--
You received this message because you are subscribed to a topic in the Google Groups "MyMediaLite" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/mymedialite/iBl89cS4WF0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to mymedialite+unsubscribe@googlegroups.com.

[Zeno Gantner]

btw I meant dense, not sparse in my last e-mail.

I would use something much simpler than MF, like linear regression based on the prior days or so.

The general use case of MML just does not fit this particular prediction problem.

Cheers,

   Z.

[Tushar Vaghasiya]

Thank you very much for your message.

Actually, I started my work on "prediction of wind power by using matrix factorization technique" for my master thesis.

I wrote this type of program in visual studio.

 

  {    // load the data

        var training_data = RatingData.Read("C:/Users/TUSHAR/Desktop/matlab csv/train.csv");

        var test_data = RatingData.Read("C:/Users/TUSHAR/Desktop/matlab csv/test.csv");

        // set up the recommender

        var recommender = new MatrixFactorization();

        recommender.Ratings = training_data;

        recommender.Regularization = 0.03f;

        recommender.NumFactors = 24;

        recommender.LearnRate = 0.01f;

        recommender.Train();

        // measure the accuracy on the test data set

        var results = recommender.Evaluate(test_data);

        Console.WriteLine("RMSE={0} MAE={1}", results["RMSE"], results["MAE"]);

        Console.WriteLine(results);

        // make a prediction for a certain user and item

        Console.WriteLine("wind speed={0} ", recommender.Predict(3652, 1));

        Console.WriteLine(recommender.DoCrossValidation());

}

could you please tell me how to select the values of Regularization, NumFactors, and LearnRate?

Thank you very much in advance,

Sincere Regards,

Tushar K. vaghasiya

[Djoels]

I've already asked this question before on this forum (roughly the same intuition as with WRMF and BPR for item recommendation applies here I guess):

• First step would be to select a "feasible" number of factors (num_factors parameter) as this will mostly influence computation time
  
• Second step: choose learn rate. Try a couple of learn rates on a logaritmic scale (0.1, 0.01, 0.001 for instance) and watch convergence of chosen evaluation metrics. (find iter option can handily be used to have an evaluation every N iterations (for instance on 5th, 10th, 15th, 20th and 25th iteration)
  
• Third step: find optimal regularization parameter. If there's only one, it's going to be very easy, if there are multiple, try a grid search approach, for instance a nested for loop of all posibilities.

PS: A very interesting read (for me) in the domain of CFD and wind speed prediction was "Employing statistical model emulation as a surrogate for CFD", a paper by Moonen and Allegrini from 2015

Op maandag 10 april 2017 15:00:09 UTC+2 schreef TUSHAR VAGHASIYA:

To unsubscribe from this group and stop receiving emails from it, send an email to mymedialite...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "MyMediaLite" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/mymedialite/iBl89cS4WF0/unsubscribe.

To unsubscribe from this group and all its topics, send an email to mymedialite...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "MyMediaLite" group.

To unsubscribe from this group and stop receiving emails from it, send an email to mymedialite...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "MyMediaLite" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/mymedialite/iBl89cS4WF0/unsubscribe.

To unsubscribe from this group and all its topics, send an email to mymedialite...@googlegroups.com.