Page 136 - Artificial Intellegence_v2.0_Class_12
P. 136

mean_ d if f _ sq  =  d if f _ sq .mean( )

            r mse_ v al =  np .sq r t( mean_ d if f _ sq )
             r etu r n r mse_ v al
                                                                   f
        p r int( " p r ed icted  v alu es ar e:  "  +  str ( [ " % .4 f "  %  i  or  i in y _ p r ed ] ) )
                                                               f
        p r int( " actu al v alu es ar e:  "  +  str ( [ " % .4 f "  %  i  or  i in y _ tr u e] ) )
        r mse_ v al =  r mse( y _ p r ed ,  y _ tr u e)
        p r int( " RM S E r r or  is:  "  +  str ( r mse_ v al) )
        Output:
        predicted values are   ' .    ', ' .    ', ' .    '
        actual values are   ' .    ', ' .    ', ' .    '
             rror is   .
        If you have a larger      value, you  ill most likely need to alter your feature or t eak your hyperparameters.


        Mean S q uar e Per centag e Er r or  ( MA PE)
         he accuracy of a forecasting technique is determined by the mean absolute percentage error   A  ). It represents the
        average of the absolute percentage errors of each entry in a dataset.  arge data sets may typically be effectively analysed
        using  A  ,  hich requires that dataset values should be other than  ero.
         A   is significant because it may assist a company in creating more precise pro ections for upcoming pro ects.

        MAPE = (1/n) * Σ(|actual value — predicted value | / |actual value|) * 100
         ince   A    displays  the  error  numbers  as  percentages,  it  is  simple  to  comprehend.   or  instance,  a   A    of
        indicates  a       difference  bet een  the  actual  and  pro ected  values.  Additionally,   hen  using  absolute  percentage
        errors, the problem of positive and negative errors cancelling each other out is eliminated.  he forecast is more accurate
         ith smaller values of  A  .


        MA PE
         he  A   penali es negative errors  ith greater intensity than positive ones.  o, it  ill choose a method  hose values
        are by default too lo   hen comparing the accuracy of prediction methods.

        H yper par ameter s
         yperparameters are parameters  hose values govern the learning process.  hey also determine the values of model
        parameters learned by a learning algorithm.  hey are 'top level' parameters that regulate the learning process and the
        model parameters that come from it, as the prefi  'hyper' suggests.  ince the model cannot modify its values during
        learning training, hyperparameters are said to be e ternal to the model.  ome e amples of hyperparameters are
        •    he ratio of train test split
        •    he optimisation algorithms' learning rate  e.g. gradient descent)
        •   In a neural net ork, the activation function selected  e.g.  igmoid,  e  ,  anh)
        •    he loss function that the model  ill employ
        •   A neural net ork's number of hidden layers
        •    he number of iterations  epochs) required to train a neural net ork

        •   A clustering task's number of clusters







                      Touchpad Artificial Intelligence (Ver. 2.0)-XII
   131   132   133   134   135   136   137   138   139   140   141