Commom Calibration methods for multivariate calibration

This is a Python library for dealing with multivariate calibration, e.g., Near infrared spectra regression and classification tasks.

Installation

Use the package manager pip to install pynir.

pip install pynir

In addition, we have also provide an online version at link

Usage

Simulata NIR spectra (spc) and reference values (conc)

from pynir.utils import simulateNIR

spc, conc = simulateNIR()

Regression

from pynir.utils import simulateNIR
from pynir.Calibration import pls

# estabilish PLS model
n_components = 10
plsModel = pls(n_components = n_components)
plsModel.fit(X,y)

yhat = plsModel.predict(X)

Classification

# simulate NIR data
from pynir.utils import simulateNIR
from pynir.Calibration import plsda

nclass = 4
X,y,wv = simulateNIR(nSample=200,n_components=10,refType=nclass, noise=1e-5)

# estabilish PLS model
n_components = 10
plsdaModel = plsda(n_components = n_components)
plsdaModel.fit(X,y)

yhat = plsdaModel.predict(X)

Feature selection

# Feature selection
from pynir.utils import simulateNIR
from pynir.Calibration import pls
from pynir.FeatureSelection import MCUVE

# simulate NIR data
X,y,wv = simulateNIR(nSample=200,n_components=10,noise=1e-5)

mcModel = MCUVE(X, y, n_components, nrep=nrep).fit()
featureSelected_MC_UVE = mcModel.featureRank[:nSel]

Outlier dection

import numpy as np
import matplotlib.pyplot as plt

from pynir.utils import simulateNIR

from pynir.OutlierDection import outlierDection_PLS

# simulate NIR data
X,y,wv = simulateNIR(nSample=200,n_components=10,noise=1e-5)

ODModel = outlierDection_PLS(ncomp=3)
Q, Tsq, Q_conf, Tsq_conf, idxOutlier = ODModel.fit(X, y).detect(X,y)
ODModel.plot_HotellingT2_Q(Q, Tsq, Q_conf, Tsq_conf)

Calibration Transfer

from pynir.utils import simulateNIR_calibrationTransfer
from pynir.Calibration import pls, regresssionReport
from pynir.CalibrationTransfer import PDS,SST, BS
import matplotlib.pyplot as plt
import numpy as np

# Simulate NIR spectra for calibration transfer
nSample = 100
X1, X2, y, wv = simulateNIR_calibrationTransfer(nSample=nSample,n_components=10,shifts=5e1)
idxTrain,idxTest = train_test_split(np.arange(nSample),test_size=0.6)
idxTransfer,idxTest = train_test_split(idxTest,test_size=0.5)


# Multivariate calibration
n_components = 7
plsModel1 = pls(n_components=n_components).fit(X1[idxTrain,:], y[idxTrain])

yhat1 = plsModel1.predict(X1[idxTest,:],n_components=n_components)
yhat2= plsModel1.predict(X2[idxTest,:],n_components=n_components)

fig, ax = plt.subplots(2,sharex=True,figsize=(8,16))
plsModel1.plot_prediction(y[idxTest], yhat1,title = "First", ax = ax[0])
plsModel1.plot_prediction(y[idxTest], yhat2, title= "Second", ax = ax[1])

# Calibration transfer on spectra
## PDS
X2_PDS = PDS(halfWindowSize=3).fit(X1[idxTransfer,:], X2[idxTransfer,:]).transform(X2[idxTest,:])
yhat2_PDS= plsModel1.predict(X2_PDS,n_components=n_components)
plsModel1.plot_prediction(y[idxTest], yhat2_PDS, title= "PDS")

fig, ax = plt.subplots()
ax.plot(wv, np.transpose(X2_PDS))


## SST
X2_SST = SST(n_components=n_components).fit(X1[idxTransfer,:], X2[idxTransfer,:]).transform(X2[idxTest,:])
yhat2_SST= plsModel1.predict(X2_SST,n_components=n_components)
plsModel1.plot_prediction(y[idxTest], yhat2_SST, title= "SST")

fig, ax = plt.subplots()
ax.plot(wv, np.transpose(X2_SST))


# Calibration transfer on prediction
## BS
yhat2_BS = BS().fit(yhat1, yhat2).transform(yhat2)
plsModel1.plot_prediction(y[idxTest], yhat2_BS, title= "BS")

Calibration Enhancement

from pynir.utils import simulateNIR_calibrationTransfer
from pynir.Calibration import pls, regresssionReport
from pynir.CalibrationTransfer import NS_PFCE,SS_PFCE,FS_PFCE,MT_PFCE
import matplotlib.pyplot as plt
import numpy as np

from sklearn.model_selection import train_test_split

import time


nSample = 100
X1, X2, y, wv = simulateNIR_calibrationTransfer(nSample=nSample,n_components=10,shifts=5e1)
idxTrain,idxTest = train_test_split(np.arange(nSample),test_size=0.6)
idxTransfer,idxTest = train_test_split(idxTest,test_size=0.5)


n_components = 7
plsModel1 = pls(n_components=n_components).fit(X1[idxTrain,:], y[idxTrain])



## PFCE
thres = 0.98
constrType = 1

tic = time.time()
b1 = plsModel1.model['B'][:,-1]
NS_PFCE_model = NS_PFCE(thres=thres, constrType=constrType).fit(X1[idxTransfer,:],X2[idxTransfer,:],b1)
yhat2_NS_PFCE = NS_PFCE_model.transform(X2[idxTest,:])
plsModel1.plot_prediction(y[idxTest], yhat2_NS_PFCE, title= "NS-PFCE")

fig, ax = plt.subplots()
ax.plot(wv, NS_PFCE_model.b2.x[1:])
ax.set_xlabel("wavelength (nm)")
ax.set_ylabel("Regression Coefficients")
ax.set_title("NS-PFCE")
print("cost {:.2f} seconds for NS-PFCE".format(time.time()-tic))


tic = time.time()
b1 = plsModel1.model['B'][:,-1]
SS_PFCE_model = SS_PFCE(thres=thres, constrType=constrType).fit(X2[idxTransfer,:],y[idxTransfer],b1)
yhat2_SS_PFCE = SS_PFCE_model.transform(X2[idxTest,:])
plsModel1.plot_prediction(y[idxTest], yhat2_SS_PFCE, title= "SS-PFCE")

fig, ax = plt.subplots()
ax.plot(wv, SS_PFCE_model.b2.x[1:])
ax.set_xlabel("wavelength (nm)")
ax.set_ylabel("Regression Coefficients")
ax.set_title("SS-PFCE")
print("cost {:.2f} seconds for SS-PFCE".format(time.time()-tic))


tic = time.time()
b1 = plsModel1.model['B'][:,-1]
FS_PFCE_model = FS_PFCE(thres=thres, constrType=constrType).fit(X1[idxTransfer,:],X2[idxTransfer,:],y[idxTransfer],b1)
yhat2_FS_PFCE = FS_PFCE_model.transform(X2[idxTest,:])
plsModel1.plot_prediction(y[idxTest], yhat2_FS_PFCE, title= "FS-PFCE")

fig, ax = plt.subplots()
ax.plot(wv, FS_PFCE_model.b2.x[1:])
ax.set_xlabel("wavelength (nm)")
ax.set_ylabel("Regression Coefficients")
ax.set_title("FS-PFCE")
print("cost {:.2f} seconds for FS-PFCE".format(time.time()-tic))


tic = time.time()
b1 = plsModel1.model['B'][:,-1]
MT_PFCE_model = MT_PFCE(thres=thres, constrType=constrType)
MT_PFCE_model.fit([X1[idxTrain,:],X2[idxTransfer,:]],(y[idxTrain],y[idxTransfer]),b1)
yhat1_MT_PFCE = MT_PFCE_model.transform(X1[idxTest,:],0) # task 1
yhat2_MT_PFCE = MT_PFCE_model.transform(X2[idxTest,:],1) # task 2

fig, ax = plt.subplots(2,sharex=True,figsize=(8,16))
plsModel1.plot_prediction(y[idxTest], yhat1_MT_PFCE, title= "MT-PFCE_First", ax= ax[0])
plsModel1.plot_prediction(y[idxTest], yhat2_MT_PFCE, title= "MT-PFCE_Second", ax= ax[1])

fig, ax = plt.subplots()
ax.plot(wv, MT_PFCE_model.B.x.reshape(2,-1)[:,1:].transpose())
ax.set_xlabel("wavelength (nm)")
ax.set_ylabel("Regression Coefficients")
ax.set_title("MT-PFCE")
print("cost {:.2f} seconds for MT-PFCE".format(time.time()-tic))

Demon

First, execute

git clone https://github.com/JinZhangLab/pynir.git
cd ./pynir/examples

Then, execute code in your python coding environment or just in terminal as follows:

python Demo1_SimulateNIR.py
python Demo2_Regression.py
python Demo3_Binary_Classification.py
python Demo4_Multiclass_Classification.py
python Demo5_dataPreprocessing.py
python Demo6_outierDection.py
python Demo7_FeatureSelection_oneStep
python Demo8_FeatureSelection_multiSteps.py
python Demo9_calibrationTransfer.py
python Demo10_calibrationTransfer_PFCE_simulateNIR.py
python Demo11_calibrationTransfer_PFCE_Tablet.py
python Demo12_calibrationTransfer_PFCE_Corn.py

Ref

• Zhang, J.; Cui, X. Y.; Cai, W. S.; Shao, X. G., A variable importance criterion for variable selection in near-infrared spectral analysis. Sci. China Chem. 2018, 62, 271-79.link

• Zhang J., Li B. Y., Hu Y., Zhou L. X., Wang G. Z., Guo G., Zhang Q. H., Lei S. C., Zhang A. H. A parameter-free framework for calibration enhancement of near-infrared spectroscopy based on correlation constraint [J]. Analytica Chimica Acta, 2021, 1142: 169-178. link

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

License

MIT