DEEP LEARNING

karrarenas1981
karrarenas1981 New Altair Community Member
edited November 2024 in Community Q&A
HELLO  everyone please if someone can help me to find the solution for my question 

i use deep learning model in rapidminer but the problem when i get the results and save it in excel sheet and do my work on it till now i am good 

but when i try apply same model again with same deep learning i get another results not same first one with out any change everything

also when i try to do same thing and apply third time i get other results different than the first and second apply 

please can i know why and how i will get the same result in same apply with out change 

thank you a lot  
Tagged:

Answers

  • jacobcybulski
    jacobcybulski New Altair Community Member
    edited July 2020
    I assume that you are using a built-in Deep Learning operator. If you are using a Deep Learning extension, you will find a similar solution. There are two issues that you may need to deal with.
    First is to do with splitting data for training and validation / testing. The Split Data operator uses a random process so that your data set is sub-divided into partitions in an unexpected way. However, if you select Advanced Parameters option, you will see a "use local random seed" option which allows you to "fix" the randomness so that it always splits data in the same way - this can be achieved by entering a specific number as a "local random seed" (any number, e.g. 2020).
    The second issue is with the Deep Learning operator, which also uses a random process in the neural network learning. Again you can "fix" it by selecting a "reproducible" option, which reveals a "use local random seed" option, which can be selected and which allows you to define a random seed (e.g. 2020), so that your network will always learn from your data in exactly the same way.
    Jacob
  • karrarenas1981
    karrarenas1981 New Altair Community Member
    you are the best thank you a lot i did you are right 
    but last thing why i use 2020 and what the benefit for this for example why not another value more or less what will be different 

    please explain to me this value thank you alot 
  • jacobcybulski
    jacobcybulski New Altair Community Member
    The value is arbitrary, so I suggested 2020, but any value will do. The idea of a specific value is that it is used as a random seed, i.e. it will initiate generation of a specific sequence of random numbers to guide all random processes in RapidMiner, e.g. selecting examples for a training / testing data partition. If the next time you (or your colleague) will use exactly the same random seed, you (they) will get exactly the same results.