def compare_models(X_stand, y_fromStand, modelsToCompare):
# -------------------------------------------------------
# Initialize comparison relevant parameters
model = lm.LinearRegression()
K1 = 10
K2 = 10
K3 = 10 # Number of total comparison loops
modelErrors = np.zeros((K3, len(modelsToCompare)))
for i in range(K3):
# -------------------------------------------------------
# Compute error for the regular model
xIn, yIn = X_stand, y_fromStand
modelErrors[i, 0] = twoLevelCV_single(xIn, yIn, model, K1, K2)
# -------------------------------------------------------
# Compute error for the 6 PCA model
xIn, yIn = X_stand, y_fromStand
modelErrors[i, 1] = twoLevelCV_single_PCA(xIn, yIn, model, K1, K2)
# -------------------------------------------------------
# Compute error for the added features model
xIn, yIn = x_add_features(X_stand, y_fromStand)
modelErrors[i, 2] = twoLevelCV_single(xIn, yIn, model, K1, K2)
# -------------------------------------------------------
# Compute error for the polynomial regression model
xIn, yIn = x_tilda_poly(X_stand, y_fromStand)
modelErrors[i, 3] = twoLevelCV_single(xIn, yIn, model, K1, K2)
# -------------------------------------------------------
# Compute error for the transformed features model
xIn, yIn = x_tilda_transform(X_stand, y_fromStand)
modelErrors[i, 4] = twoLevelCV_single(xIn, yIn, model, K1, K2)
# -------------------------------------------------------
# Compute error for the chosen features model
features = np.array([1, 2])
xIn, yIn = x_tilda_downSample(xIn, yIn, features)
modelErrors[i, 5] = twoLevelCV_single(xIn, yIn, model, K1, K2)
# -------------------------------------------------------
# Compute error for the transfomrmed + PCA features model
xIn, yIn = x_tilda_transform(X_stand, y_fromStand)
modelErrors[i, 6] = twoLevelCV_single_PCA(xIn, yIn, model, K1, K2)
# MSE calculation - for plots
# modelErrorsAvg = np.mean(modelErrors, axis = 0)
# RMSE calculation - for plots
modelErrorsAvg = np.sqrt(np.mean(modelErrors, axis=0))
return modelErrorsAvg
Created: 13.11.2020
"""
import matplotlib.pyplot as plt
import numpy as np
import torch
from sklearn import model_selection
import sklearn.linear_model as lm
from scipy import stats
from ANN_functions import *
from concNoZero_config import *
from featureTransform import x_add_features
from regularization import rlr_validate, regmultinominal_regression
xIn,yIn = x_add_features(X_stand, y_fromStand)
M = xIn.shape[1]
attributeNames.append('Xf1')
attributeNames.append('Xf2')
attributeNames.append('Xf3')
classNames = classNames.tolist()
#%%
# Unbalanced Dataset
print("Observations of Low Concrete = {}".format(np.sum(y_class.squeeze()==0)))
print("Observations of Medium Concrete = {}".format(np.sum(y_class.squeeze()==1)))
print("Observations of High Concrete = {}".format(np.sum(y_class.squeeze()==2)))
# BASELINE CLASSIFICATION MODEL
baseline_class = np.array((np.sum(y_class.squeeze()==0), np.sum(y_class.squeeze()==1), np.sum(y_class.squeeze()==2)))
baseline_model_prediction = np.argmax(baseline_class)*np.ones(y_class.shape[0])
def correlated_ttest(r, rho, alpha=0.05):
rhat = np.mean(r)
shat = np.std(r)
J = len(r)
sigmatilde = shat * np.sqrt(1 / J + rho / (1 - rho))
CI = st.t.interval(1 - alpha, df=J - 1, loc=rhat,
scale=sigmatilde) # Confidence interval
p = 2 * st.t.cdf(-np.abs(rhat) / sigmatilde, df=J - 1) # p-value
return p, CI
#_______CREATE DATASET WITH ADDED FEATURES_______
xIn, yIn = x_add_features(X_stand, y_class)
# Initialize 2 layer CV parameters
K1 = 5
K2 = 5
# Values of lambda
lambdas = np.logspace(-5, 5, 20)
# Range of hidden units
hidden_units = np.array((1, 3, 6, 8, 11, 15))
# Parameters for ANN training part
CV_ann = 2
n_replicates = 1
max_iter = 15000
tolerance = 1e-7