ProcessPLS 1.4

Implementation of ProcessPLS in Python

ProcessPLS

An Implementation of ProcessPLS in Python

Code Writter

Implementation by Sin Yong Teng. Radboud University Nijmegen, the Netherlands.

Implementation

In this code implementation, the sklearn syntax is used. Furthermore, the ProcessPLS algorithm has been made to be represented in directed graphs data structure. This allows for more flexibility to be used with graph theory routines.

Functions

Install the library

pip install processPLS

Get the data

from processPLS.model import *
from processPLS.data import *
X,Y,matrix=ValdeLoirData() #Get the data conviniently

Alternatively, you can import the data yourself like this:

df=pd.read_csv(r'.\ValdeLoirData.csv')
df=df.drop(columns=df.columns[0])
smell_at_rest=df.iloc[:,:5]
view=df.iloc[:,5:8]
smell_after_shaking=df.iloc[:,8:18]
tasting=df.iloc[:,18:27]
global_quality=df.iloc[:,27]

X={
'Smell at Rest':smell_at_rest,
"View":view,
"Smell after Shaking":smell_after_shaking,
"Tasting":tasting,
}

Y={"Global Quality":global_quality}

matrix = pd.DataFrame(
[
[0,0,0,0,0], 
[1,0,0,0,0],
[1,1,0,0,0],
[1,1,1,0,0],
[1,1,1,1,0],
],
index=list(X.keys())+list(Y.keys()),
columns=list(X.keys())+list(Y.keys())
)

Call and Fit the Process PLS model

import matplotlib.pyplot as plt
model = ProcessPLS()
model.fit(X,Y,matrix)
model.plot
plt.show()

Main Function Arguments

Process_PLS(cv=RepeatedKFold(n_splits=5,n_repeats=2,random_state=999),scoring='neg_mean_squared_error',max_lv=30,overwrite_lv=False,inner_forced_lv=None,outer_forced_lv=None,name=None)

'''
This function sets up the processPLS model.

cv= cross validation method  (follows sklearn syntax)

scoring= loss function/ scoring function (follows sklearn syntax)

max_lv= maximum numbers of latent variable (lv) for all SIMPLS models within ProcessPLS

overwrite_LV= (True/False) A boolean to set whether inner_forced_lv and outer_forced_lv should be used instead of automatically selecting latent variables

inner_forced_lv= (dict) a specific key value combination of number of LVs to forced into the inner model. Argument overwrite_LV must be set to True for this to be used. Example input:
 inner_forced_lv={
  'Smell at Rest':None,
  "View":3,
  "Smell after Shaking":6,
  "Tasting":8,
  "Global Quality":13
  }

  inner_forced_lv= (dict) a specific key value combination of number of LVs to forced into the outer model. Argument overwrite_LV must be set to True for this to be used. Example input:

  outer_forced_lv={
  'Smell at Rest':3,
  "View":3,
  "Smell after Shaking":2,
  "Tasting":5,
  "Global Quality":3
  }

name: (string) Optional name of model.

'''

ValdeLoirData(original=False)

'''
This function gets the data for Valde Loir Dataset

original==False:  The function returns X (dataframe in dict), Y (dataframe dict), and matrix (dataframe). matrix is the adjacency matrix for the graph connections.

original==True:  The function returns the raw data (dataframe) with both X and Y combined within


'''

Inference/ Prediction for New Data

y_pred= model.predict(Xnew)

Reproducibility

This implementation provides exactly the same output as the MATLAB version of ProcessPLS.

Reference to Original Paper:

van Kollenburg, G., Bouman, R., Offermans, T., Gerretzen, J., Buydens, L., van Manen, H.J. and Jansen, J., 2021. Process PLS: Incorporating substantive knowledge into the predictive modelling of multiblock, multistep, multidimensional and multicollinear process data. Computers & Chemical Engineering, 154, p.107466.

For MATLAB Implementation, see this repository written by Tim Offermans. https://gitlab.science.ru.nl/toffermans/matlab-process-pls/-/tree/main/