diff --git a/src/Class_Mod/DATA_HANDLING.py b/src/Class_Mod/DATA_HANDLING.py
index 56af54af8fa7f40d237e2febd6972590a1cb72d1..84cb46e0ec254a9526cf0eca39333f405b198655 100644
--- a/src/Class_Mod/DATA_HANDLING.py
+++ b/src/Class_Mod/DATA_HANDLING.py
@@ -80,4 +80,33 @@ def Snv(X):
return pd.DataFrame(c.T, index=X.index, columns= X.columns)
def Non(X):
- return X
\ No newline at end of file
+ return X
+
+
+######################################## Cross val split ############################
+########################## Kennard-Stone split method for generating CV-folds ######################################
+def CV(x, y, n_folds:int):
+ test_folds = {}
+ folds_name = [f'cv{i+1}' for i in range(n_folds)]
+ kf = ks.KFold(n_splits=3, device='cpu')
+
+ for i in range(n_folds):
+ d = []
+ for _, i_test in kf.split(x, y):
+ d.append(i_test)
+ test_folds[folds_name[i]] = d[i]
+ return test_folds
+
+################################### Cross validate a model and return predictions/fold and samples idx #######################################################
+def cross_val_predictor(model, x, y, nfolds):
+ x = np.array(x)
+ y = np.array(y)
+
+ yp = {}
+ folds = CV(x=x, y=y, n_folds=3)### Test index
+ key = list(folds.keys())
+
+ for i in range(nfolds):
+ model.fit(np.delete(x, folds[key[i]], axis=0), np.delete(y, folds[key[i]], axis=0))
+ yp[key[i]] = model.predict(x[folds[key[i]]]) #### predictions/fold
+ return yp, folds
\ No newline at end of file
diff --git a/src/Class_Mod/Evaluation_Metrics.py b/src/Class_Mod/Evaluation_Metrics.py
new file mode 100644
index 0000000000000000000000000000000000000000..ebe94c0fa613913c426ce373bb649f7e5315ae8b
--- /dev/null
+++ b/src/Class_Mod/Evaluation_Metrics.py
@@ -0,0 +1,56 @@
+from Packages import *
+
+class metrics:
+ def __init__(self, c:Optional[float] = None, cv:Optional[List] = None, t:Optional[List] = None, method = 'regression')-> pd.DataFrame:
+ phase = [c, cv, t]
+ index = np.array(["train", "cv", "test"])
+ notnone = [i for i in range(3) if phase[i] != None]
+ met_index = index[notnone]
+ methods = ['regression', 'classification']
+ perf = {}
+ for i in notnone:
+ if method == 'regression':
+ perf[index[i]] = metrics.reg_(phase[i][0], phase[i][1])
+
+ elif method == 'classification':
+ perf[index[i]] = metrics.class_(phase[i][0], phase[i][1])
+
+ if notnone == 1:
+ self.ret = perf.T
+ else:
+ self.ret = pd.DataFrame(perf).T
+
+ @staticmethod
+ def reg_(meas, pred):
+ meas = np.array(meas)
+ pred = np.array(pred)
+ xbar = np.mean(meas) # the average of measured values
+ e = np.subtract(meas , pred)
+ e2 = e**2# the squared error
+
+ # Sum of squared:
+ # TOTAL
+ sst = np.sum((meas - xbar)**2)
+ # RESIDUAL
+ ssr = np.sum(e2)
+ # REGRESSION OR MODEL
+ ssm = np.sum(pred - xbar)
+
+
+ # Compute statistical metrics
+ metr = {}
+ metr['r'] = np.corrcoef(meas, pred)[0, 1]
+ metr['r2'] = 1-ssr/sst
+ metr['rmse'] = np.sqrt(np.mean(e2))
+ metr['mae'] = np.mean(np.abs(e2))
+ metr['rpd'] = np.std(meas)/np.sqrt(np.mean(e2))
+ metr['rpiq'] = (np.quantile(meas, .75) - np.quantile(meas, .25))/np.sqrt(np.mean(e2))
+ return metr
+
+ @staticmethod
+ def class_(meas, pred):
+ pass
+
+ @property
+ def scores_(self):
+ return self.ret
\ No newline at end of file
diff --git a/src/Class_Mod/PLSR_.py b/src/Class_Mod/PLSR_.py
index 1ccd3d08ddf767e1a7a1a917f7c219d701849a33..062f17026d3ee8a6db86b869537b6989b39cc729 100644
--- a/src/Class_Mod/PLSR_.py
+++ b/src/Class_Mod/PLSR_.py
@@ -1,6 +1,6 @@
from Packages import *
from Class_Mod.Miscellaneous import *
-from Class_Mod.Regression_metrics import metrics
+from Class_Mod.Evaluation_Metrics import metrics
class PinardPlsr:
def __init__(self, x_train, y_train, x_test, y_test):
@@ -45,21 +45,6 @@ class PinardPlsr:
def model_(self):
return self.trained
- @property
- def metrics_(self):
- metc = metrics(self.y_train, self.yc)
- metc = metc.evaluate_
-
- metcv = metrics(self.y_train, self.ycv)
- metcv = metcv.evaluate_
-
- mett = metrics( self.y_test, self.yt)
- mett = mett.evaluate_
-
- met = pd.concat([metc, metcv, mett], axis = 0)
- met.index = ['calib','cv','test']
- return met
-
@property
def pred_data_(self):
diff --git a/src/Class_Mod/PLSR_Preprocess.py b/src/Class_Mod/PLSR_Preprocess.py
index 72aa9841093da02e3bafc84371f9cbd3ce4e12f5..7904ef6d8033a6fc23acd0ac0e7703a60987acef 100644
--- a/src/Class_Mod/PLSR_Preprocess.py
+++ b/src/Class_Mod/PLSR_Preprocess.py
@@ -94,21 +94,7 @@ class PlsProcess:
@property
def model_(self):
return self.model
- @property
- def metrics_(self):
- metc = metrics(self.y_train, self.yc)
- metc = metc.evaluate_
-
- metcv = metrics(self.y_train, self.ycv)
- metcv = metcv.evaluate_
-
- mett = metrics( self.y_test, self.yt)
- mett = mett.evaluate_
-
- met = pd.concat([metc, metcv, mett], axis = 0)
- met.index = ['calib','cv','test']
- return met
-
+
@property
def pred_data_(self):
return self.yc, self.ycv, self.yt
\ No newline at end of file
diff --git a/src/Class_Mod/Regression_metrics.py b/src/Class_Mod/Regression_metrics.py
deleted file mode 100644
index 5ec8cacb36eb7eecd515311aa3ec0f2f4120be24..0000000000000000000000000000000000000000
--- a/src/Class_Mod/Regression_metrics.py
+++ /dev/null
@@ -1,31 +0,0 @@
-from Packages import *
-class metrics:
- def __init__(self, meas, pred):
- if isinstance(meas, pd.DataFrame) or isinstance(meas, pd.Series):
- self.meas = np.array(meas).reshape(-1)
- self.pred = np.array(pred).reshape(-1)
-
- @property
- def evaluate_(self):
- xbar = np.mean(self.meas) # the average of measured values
- e = np.subtract(self.meas, self.pred)
- e2 = e**2# the squared error
-
- # Sum of squared:
- # TOTAL
- sst = np.sum((self.meas- xbar)**2)
- # RESIDUAL
- ssr = np.sum(e2)
- # REGRESSION OR MODEL
- ssm = np.sum(self.pred - xbar)
-
-
- # Compute statistical metrics
- metr = pd.DataFrame()
- metr['r'] = [np.corrcoef(self.meas, self.pred)[0,1]]
- metr['r2'] = [1-ssr/sst]
- metr['rmse'] = [np.sqrt(np.mean(e2))]
- metr['mae'] = [np.mean(np.abs(e2))]
- metr['rpd'] = [np.std(self.meas)/np.sqrt(np.mean(e2))]
- metr['rpiq'] = [(np.quantile(self.meas,.75)-np.quantile(self.meas,.25))/np.sqrt(np.mean(e2))]
- return metr.round(3)
\ No newline at end of file
diff --git a/src/Class_Mod/SK_PLSR_.py b/src/Class_Mod/SK_PLSR_.py
index 7b9de1cbf2f1e776d91b97d700ab58b2893ec079..f08082f99fade1759f9837fb8ae059742f0456a3 100644
--- a/src/Class_Mod/SK_PLSR_.py
+++ b/src/Class_Mod/SK_PLSR_.py
@@ -1,64 +1,121 @@
from Packages import *
from Class_Mod.Miscellaneous import *
-from Class_Mod.Regression_metrics import metrics
-
-
-
+from Class_Mod.Evaluation_Metrics import metrics
+from Class_Mod.DATA_HANDLING import Snv
class PlsR:
+ SCORE = 100000000
+
def __init__(self, x_train, y_train, x_test, y_test):
+ self.PLS_params = {}
+ a = [0, 1, 2]
+ if min(a)==0:
+ b = [0]
+ elif min(a)==1:
+ b= [0,1]
+ elif min(a) ==2:
+ b = [0, 1, 2]
+
+ self.PLS_params['Preprocess'] = {'Scatter':hp.choice('Scatter',['Snv', None]),
+ 'window_length_sg':hp.choice('window_length_sg', [9, 13, 17, 21]),
+ 'polyorder_sg':hp.choice('polyorder_sg',a),
+ 'deriv_sg':hp.choice('deriv_sg', b)}
+
+ self.PLS_params['n_components'] = hp.choice("n_components", list(np.arange(1,21)))
+
self.x_train = x_train
self.x_test = x_test
self.y_train = y_train
self.y_test = y_test
-
- self.trained = PLSRegression(n_components= self._optimize(), scale = False)
- self.trained.fit(self.x_train, self.y_train)
+ self.p = self.x_train.shape[1]
+
+ trials = Trials()
+ best_params = fmin(fn=self.objective,
+ space=self.PLS_params,
+ algo=tpe.suggest, # Tree of Parzen Estimators’ (tpe) which is a Bayesian approach
+ max_evals=100,
+ trials=trials,
+ verbose=0)
+ #####################################################################################################
+ if self.best['Preprocess']['Scatter'] is None:
+ xtrain = self.x_train
+ xtest = self.x_test
+ elif self.best['Preprocess']['Scatter'] == 'Snv':
+ xtrain = Snv(self.x_train)
+ xtest = Snv(self.x_test)
+
+ x_train = savgol_filter(xtrain, window_length = self.best['Preprocess']['window_length_sg'],
+ polyorder = self.best['Preprocess']['polyorder_sg'],
+ deriv=self.best['Preprocess']['deriv_sg'])
+
+ x_test = savgol_filter(xtest, window_length = self.best['Preprocess']['window_length_sg'],
+ polyorder = self.best['Preprocess']['polyorder_sg'],
+ deriv=self.best['Preprocess']['deriv_sg'])
+
- self.yc = pd.DataFrame(self.trained.predict(self.x_train)) # make predictions on test data and assign to Y_preds variable
- self.ycv = pd.DataFrame(cross_val_predict(self.trained, self.x_train, self.y_train, cv = 3)) # make predictions on test data and assign to Y_preds variable
- self.yt = pd.DataFrame(self.trained.predict(self.x_test)) # make predictions on test data and assign to Y_preds variable
+ ######################################################################################################
+ self.trained = PLSRegression(n_components= self.best['n_components'], scale = False)
+ self.trained.fit(x_train, self.y_train)
- def _optimize(self):
- nlv = 21
- rmse = np.ones(21)
- rmse[0] = 0.002
- lv = {}
- ratio = []
- for i in range(1,nlv):
- m = PLSRegression(n_components= i, scale = False)
- ycv = cross_val_predict(m, self.x_train, self.y_train, cv = 5)
- rmse[i] = mean_squared_error(self.y_train, ycv)
+ self.yc = pd.DataFrame(self.trained.predict(x_train)) # make predictions on test data and assign to Y_preds variable
+ self.ycv = pd.DataFrame(cross_val_predict(self.trained, x_train, self.y_train, cv = 3)) # make predictions on test data and assign to Y_preds variable
+ self.yt = pd.DataFrame(self.trained.predict(x_test)) # make predictions on test data and assign to Y_preds variable
+ #######################################################################################################
+
+ def objective(self, params):
+ ws = params['Preprocess']['window_length_sg']
+ po = params['Preprocess']['polyorder_sg']
+ dr = params['Preprocess']['deriv_sg']
- ratio.append(((rmse[i-1]-rmse[i])/rmse[i-1])*100)
- return np.argmax(ratio)+1
+ if params['Preprocess']['Scatter'] is None:
+ xtrain = self.x_train
+ xtest = self.x_test
+ elif params['Preprocess']['Scatter'] == 'Snv':
+ xtrain = Snv(self.x_train)
+ xtest = Snv(self.x_test)
- ################################################################################################################
+ x_train = savgol_filter(xtrain, window_length = params['Preprocess']['window_length_sg'],
+ polyorder = params['Preprocess']['polyorder_sg'],
+ deriv=params['Preprocess']['deriv_sg'])
+
+ x_test = savgol_filter(xtest, window_length = params['Preprocess']['window_length_sg'],
+ polyorder = params['Preprocess']['polyorder_sg'],
+ deriv=params['Preprocess']['deriv_sg'])
+
+ m = PLSRegression( n_components= params['n_components'], scale = False )
+ m.fit(x_train, self.y_train)
- ################################################################################################################
+ yc = m.predict(x_train)
+ ycv = cross_val_predict( m, x_train, self.y_train, cv = 5)
+ yt = m.predict(x_test)
+
+ rmsec = mean_squared_error(self.y_train, yc)
+ rmsecv = mean_squared_error(self.y_train, ycv)
+ rmset = mean_squared_error(self.y_test, yt)
+
+ SCORE = (rmsecv/rmset) + (rmsecv/rmsec) + (rmset/rmsec)
+ if SCORE < PlsR.SCORE:
+ PlsR.SCORE = SCORE
+ self.best = params
+ print(SCORE)
+ print(params)
+ return SCORE
+
@property
def model_(self):
return self.trained
-
- @property
- def metrics_(self):
- metc = metrics(self.y_train, self.yc)
- metc = metc.evaluate_
-
- metcv = metrics(self.y_train, self.ycv)
- metcv = metcv.evaluate_
-
- mett = metrics( self.y_test, self.yt)
- mett = mett.evaluate_
-
- met = pd.concat([metc, metcv, mett], axis = 0)
- met.index = ['calib','cv','test']
- return met
+ @property
+ def best_hyperparams(self):
+ self.best
+ self.b = {'Scatter':self.best['Preprocess']['Scatter'], 'Saitzky-Golay derivative parameters':{'polyorder':self.best['Preprocess']['polyorder_sg'],
+ 'deriv':self.best['Preprocess']['deriv_sg'],
+ 'window_length':self.best['Preprocess']['window_length_sg']}}
+ return self.b
+
@property
def pred_data_(self):
-
return self.yc, self.ycv, self.yt
\ No newline at end of file
diff --git a/src/Class_Mod/__init__.py b/src/Class_Mod/__init__.py
index 155f62559033d087dcb7c42c65e1bdaac7fb5aba..e9d334d53ab7739a5e4832d384016eaf1920b657 100644
--- a/src/Class_Mod/__init__.py
+++ b/src/Class_Mod/__init__.py
@@ -6,7 +6,7 @@ from .UMAP_ import Umap
from .DATA_HANDLING import *
from .PLSR_ import PinardPlsr
from .LWPLSR_ import LWPLSR
-from .Regression_metrics import metrics
+from .Evaluation_Metrics import metrics
from .VarSel import TpeIpls
from .Miscellaneous import resid_plot, reg_plot
from .DxReader import DxRead, read_dx
diff --git a/src/Packages.py b/src/Packages.py
index 7e8bb19e17b4ebdd58ad1d7b5ffeccaa02986807..debec3041c731c54064b7b5c1eb0f2a5ee393548 100644
--- a/src/Packages.py
+++ b/src/Packages.py
@@ -11,11 +11,13 @@ import datetime
import numpy as np
import pandas as pd
from matplotlib import colors
+from typing import Optional, List
from sklearn.preprocessing import StandardScaler, MinMaxScaler, LabelEncoder
import time
from scipy.stats import skew, kurtosis
from scipy.signal import savgol_filter, find_peaks_cwt, detrend
import scipy as sc
+import kennard_stone as ks
### Exploratory data analysis-Dimensionality reduction
from umap.umap_ import UMAP
from sklearn.decomposition import PCA, NMF
diff --git a/src/data/params/Preprocessing.json b/src/data/params/Preprocessing.json
index 821db7c9c527189fe8a3a6dee1f63a20957e807e..fac056691fb6176471327aacf46b1d55a7c9f80b 100644
--- a/src/data/params/Preprocessing.json
+++ b/src/data/params/Preprocessing.json
@@ -1 +1 @@
-{"Scatter": "Non", "Saitzky-Golay derivative parameters": {"polyorder": 0, "deriv": 0, "window_length": 23}}
\ No newline at end of file
+{"Scatter": "Snv", "Saitzky-Golay derivative parameters": {"polyorder": 2, "deriv": 1, "window_length": 15}}
\ No newline at end of file
diff --git a/src/pages/2-model_creation.py b/src/pages/2-model_creation.py
index 86013f4393622706a6af041f359fd9fff7ed584e..f22968685cfc4893c0f74aed0ba575c28048776f 100644
--- a/src/pages/2-model_creation.py
+++ b/src/pages/2-model_creation.py
@@ -210,7 +210,7 @@ if not spectra.empty and not y.empty:
json.dump(Reg.best_hyperparams, outfile)
M2.write("-- Performance metrics --")
- M2.dataframe(Reg.metrics_)
+ M2.dataframe(metrics(c = [y_train, yc], cv = [y_train, ycv], t = [y_test, yt], method='regression').scores_)
#from st_circular_progress import CircularProgress
#my_circular_progress = CircularProgress(label = 'Performance',value = 50, key = 'my performance',
# size = "medium", track_color = "black", color = "blue")