How to modify the scaling technique used in neural network ,can we specify the scaling technique to be used in training ?

[Bingsnow]

Actually i want to change the give in the input which scaling method to be used in my model ?

[sgoodlett]

There are three ways PES-Learn will scale data, and they are by default treated as hyperparameters that are optimized against the validation data set. There is a scaling of the data to a normal Gaussian distribution std, and the other two scale to a min max of 0 to 1 mm01 or -1 to 1 mm11. The input data is never scaled by mm01 in the hyperparameter search it seems. If you want to build a model and skip the hyperparameter optimization, thus selecting your own scaling method etc., you can something like:

import peslearn

input_string = ("""
               hp_maxit = 15
               training_points = 1000
               validation_points = 200
               use_pips = true
               sampling = structure_based
               """)

input_obj = peslearn.InputProcessor(input_string)

nn = peslearn.ml.NeuralNetwork("ml_data.dat", input_obj, molecule_type="A2B2C")
params = {'morse_transform': {'morse': False}, 'scale_X': {'scale_X': 'std', 'activation': 'tanh'}, 'scale_y': 'std', 'pip': {'degree_reduction': False, 'pip': True}, 'layers': (16,)}
nn.build_model(params)