npls 0.0.7

It is a Python library for the N-PLS1 regression with L1 and L2-regularization.

npls

npls - this module is a Python library for the N-PLS1 regression with L1 and L2-regularization.

Installation

Install the current version with PyPI:

pip install npls

Or from Github:

pip install https://github.com/89605502155/npls/main.zip

Usage

You can fit your own regression model. n_components - is a number of components of SVD decomposition and l2 is a parameter of L2-regularization (Ridge) and l1 is a parameter of L1-regularization (LASSO), excitation_wavelenth - is an excitation wavelenth of fluor. (if you have not excitation wavelenth you can write list with len equal first shape of X_train),emission_wavelenth is an emission wavelenth. X_train - is a 3d-array. y_train -is a vector.

from npls import npls 

model=npls(n_components=4, l2=0.09)
model.fit(X_train,y_train)
#components of svd-decomposition
w_i=model.w_i
w_k=model.w_k
#predict
y_predicted=model.predict(X_test)

If you want to use snr, you need input excitation_wavelenth and emission_wavelenth, crash_norm_name and crash_norm_value and norm_func and derivative_rang. Example:

from npls import npls 
model=npls(crash_norm_name='evklid',derivative_rang=[1],
    emission_wavelenth=np.array([1,2,3]),
    excitation_wavelenth=np.array([1,2,3]),
    crash_norm_value=9)

and with use L2 and N components:

from npls import npls 
model=npls(crash_norm_name='evklid',derivative_rang=[1],
    emission_wavelenth=np.array([1,2,3]),
    excitation_wavelenth=np.array([1,2,3]),
    crash_norm_value=9,n_components=4, l2=0.09)

Example

You can use this library with Scikit-learn library. For example, we can use GridSearchCV.

If you installed a module from PyPi, you should to import it like this: from npls import npls

If from GitHub or source: from npls import npls

from npls import npls 
from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score, make_scorer
import sklearn
from sklearn.model_selection import GridSearchCV

npls1=npls(excitation_wavelenth=[501,552],emission_wavelenth=[553,604])
#you can use many error metrics
scoring={'mse': make_scorer(mean_squared_error),'r2':'r2'}
parametrsNames={'n_components': [4],
                'l2': np.logspace(-25, 25,num = 51),
                'l1': np.logspace(-25, 25,num = 51)
                }

gridCought=GridSearchCV(npls1, parametrsNames, cv=5, 
    scoring=scoring,refit='r2',return_train_score=True)
gridCought.fit(X_train,y_train)
#errors
r2_p=gridCought.score(X_test.copy(), y_test.copy())
mse_cv=gridCought.cv_results_[ "mean_test_mse" ]
mse_c=gridCought.cv_results_[ "mean_train_mse" ]
r2_cv=gridCought.cv_results_[ "mean_test_r2" ]
r2_c=gridCought.cv_results_[ "mean_train_r2" ]