Skip to content
Snippets Groups Projects
Commit 2d744645 authored by Jan Habscheid's avatar Jan Habscheid
Browse files

Added Error over R

parent ec4bc5a4
Branches
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
Showing
with 247 additions and 26 deletions
04_TrainModels.py
+ 71
− 25
View file @ 2d744645
import numpy as np
import matplotlib.pyplot as plt
# from models.RegressionModels import RegressionModels
from models.LinearRegression import LinearRegressionModel
......@@ -25,41 +26,58 @@ for model_name in models:
# Search for the best number of POD modes concerning the rmse (root mean square error)
# R = np.linspace(1, 25, 25, dtype=int)
# R = np.linspace(1, 21, 21, dtype=int)
R = np.linspace(1, 21, 3, dtype=int)
R = np.linspace(1, 21, 21, dtype=int)
mae_velocity_best, mae_pressure_best = model.mae_test(model_velocity, model_pressure)
mape_velocity_best, mape_pressure_best = model.mape_test(model_velocity, model_pressure)
rmse_velocity_best, rmse_pressure_best = model.rmse_test(model_velocity, model_pressure)
max_error_velocity_best, max_error_pressure_best = model.max_error_test(model_velocity, model_pressure)
# mae_velocity_best, mae_pressure_best = model.mae_test(model_velocity, model_pressure)
# mape_velocity_best, mape_pressure_best = model.mape_test(model_velocity, model_pressure)
# rmse_velocity_best, rmse_pressure_best = model.rmse_test(model_velocity, model_pressure)
# max_error_velocity_best, max_error_pressure_best = model.max_error_test(model_velocity, model_pressure)
best_model_velocity, best_model_pressure = model_velocity, model_pressure
R_best = model.elbow_point(scaled=True)
L2_velocity_best, L2_pressure_best = model.relative_L2(model_velocity, model_pressure)
print(f'Finding the optimal number of nodes for model {model_name}')
print(f'Searching for best R in {R}')
L2_velocity_vec, L2_pressure_vec = [], []
for R_ in R:
print(f'Current R: {R_}')
X_r_velocity, X_r_pressure = model.POD(R_velocity=R_, R_pressure=R_)
model_velocity, model_pressure = model.train(X_r_velocity, X_r_pressure)
mae_velocity_, mae_pressure_ = model.mae_test(model_velocity, model_pressure)
mape_velocity_, mape_pressure_ = model.mape_test(model_velocity, model_pressure)
rmse_velocity_, rmse_pressure_ = model.rmse_test(model_velocity, model_pressure)
max_error_velocity_, max_error_pressure_ = model.max_error_test(model_velocity, model_pressure)
L2_velocity_, L2_pressure_ = model.relative_L2(model_velocity, model_pressure)
L2_velocity_vec.append(L2_velocity_)
L2_pressure_vec.append(L2_pressure_)
# mae_velocity_, mae_pressure_ = model.mae_test(model_velocity, model_pressure)
# mape_velocity_, mape_pressure_ = model.mape_test(model_velocity, model_pressure)
# rmse_velocity_, rmse_pressure_ = model.rmse_test(model_velocity, model_pressure)
# max_error_velocity_, max_error_pressure_ = model.max_error_test(model_velocity, model_pressure)
if rmse_velocity_ < rmse_velocity_best:
if L2_velocity_ < L2_velocity_best:
best_model_velocity = model_velocity
mae_velocity_best = mae_velocity_
mape_velocity_best = mape_velocity_
rmse_velocity_best = rmse_velocity_
max_error_velocity_best = max_error_velocity_
L2_velocity_best = L2_velocity_
L2_pressure_best = L2_pressure_
R_best = (R_, R_best[1])
if rmse_pressure_ < rmse_pressure_best:
if L2_pressure_ < L2_pressure_best:
best_model_pressure = model_pressure
mae_pressure_best = mae_pressure_
mape_pressure_best = mape_pressure_
rmse_pressure_best = rmse_pressure_
max_error_pressure_best = max_error_pressure_
L2_velocity_best = L2_velocity_
L2_pressure_best = L2_pressure_
R_best = (R_best[0], R_)
# if rmse_velocity_ < rmse_velocity_best:
# best_model_velocity = model_velocity
# mae_velocity_best = mae_velocity_
# mape_velocity_best = mape_velocity_
# rmse_velocity_best = rmse_velocity_
# max_error_velocity_best = max_error_velocity_
# R_best = (R_, R_best[1])
# if rmse_pressure_ < rmse_pressure_best:
# best_model_pressure = model_pressure
# mae_pressure_best = mae_pressure_
# mape_pressure_best = mape_pressure_
# rmse_pressure_best = rmse_pressure_
# max_error_pressure_best = max_error_pressure_
# R_best = (R_best[0], R_)
model.save_model(best_model_velocity, best_model_pressure, model_velocity_file=f'{model_name}/velocity_{R_best[0]}.pkl', model_pressure_file=f'{model_name}/pressure_{R_best[1]}.pkl')
model.save_model(best_model_velocity, best_model_pressure, model_velocity_file=f'{model_name}/velocity.pkl', model_pressure_file=f'{model_name}/pressure.pkl')
......@@ -68,13 +86,41 @@ for model_name in models:
print('---------------------------------')
print(f'Model: {model_name}')
print(f'\t \t\t Velocity \t Pressure')
print(f'MAE: \t\t\t {mae_velocity_best} \t {mae_pressure_best}')
print(f'MAPE: \t\t\t {mape_velocity_best} \t {mape_pressure_best}')
print(f'RMSE: \t\t\t {rmse_velocity_best} \t {rmse_pressure_best}')
print(f'max error: \t\t {max_error_velocity_best} \t {max_error_pressure_best}')
# print(f'MAE: \t\t\t {mae_velocity_best} \t {mae_pressure_best}')
# print(f'MAPE: \t\t\t {mape_velocity_best} \t {mape_pressure_best}')
# print(f'RMSE: \t\t\t {rmse_velocity_best} \t {rmse_pressure_best}')
# print(f'max error: \t\t {max_error_velocity_best} \t {max_error_pressure_best}')
print(f'$L_2$: \t\t\t {L2_velocity_best} \t {L2_pressure_best}')
print(f'Number of POD modes: \t {R_best[0]} \t {R_best[1]}')
print(f'Elbow point: \t\t {model.elbow_point(scaled=True)[0]} \t {model.elbow_point(scaled=True)[1]}')
print('---------------------------------')
print('\n \n')
np.savez(f'Results/{model_name}/OptimalSetup.npz', Elbow_points=np.array([model.elbow_point(scaled=True)[0], model.elbow_point(scaled=True)[1]]), R_best=R_best, mae_velocity_best=mae_velocity_best, mae_pressure_best=mae_pressure_best, mape_velocity_best=mape_velocity_best, mape_pressure_best=mape_pressure_best, rmse_velocity_best=rmse_velocity_best, rmse_pressure_best=rmse_pressure_best, max_error_velocity_best=max_error_velocity_best, max_error_pressure_best=max_error_pressure_best)
\ No newline at end of file
np.savez(f'Results/{model_name}/OptimalSetup.npz', Elbow_points=np.array([model.elbow_point(scaled=True)[0], model.elbow_point(scaled=True)[1]]), R_best=R_best, L2_velocity_vec=L2_velocity_vec, L2_pressure_vec=L2_pressure_vec, R=R)#mae_velocity_best=mae_velocity_best, mae_pressure_best=mae_pressure_best, mape_velocity_best=mape_velocity_best, mape_pressure_best=mape_pressure_best, rmse_velocity_best=rmse_velocity_best, rmse_pressure_best=rmse_pressure_best, max_error_velocity_best=max_error_velocity_best, max_error_pressure_best=max_error_pressure_best)
fig, axs = plt.subplots(nrows=1, ncols=2)
fig.suptitle(f'{model_name} - Error over number of POD modes')
axs[0].plot(R, L2_velocity_vec)
axs[0].set_title('u')
axs[0].set_xlabel('R')
axs[1].plot(R, L2_pressure_vec)
axs[1].set_title('p')
axs[1].set_xlabel('R')
fig.tight_layout()
fig.savefig(f'Results/{model_name}/Error_POD_Nodes.jpg')
fig, axs = plt.subplots(nrows=1, ncols=2)
fig.suptitle(f'{model_name} - Error over number of POD modes')
axs[0].semilogy(R, L2_velocity_vec)
axs[0].set_title('u')
axs[0].set_xlabel('R')
axs[1].semilogy(R, L2_pressure_vec)
axs[1].set_title('p')
axs[1].set_xlabel('R')
fig.tight_layout()
fig.savefig(f'Results/{model_name}/Error_POD_Nodes_semilogy.jpg')
99_VisError_over_R.ipynb 0 → 100644
+ 159
− 0
View file @ 2d744645
%% Cell type:code id: tags:
``` python
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
```
%% Cell type:code id: tags:
``` python
from models.LinearRegression import LinearRegressionModel
from models.GaussianProcessRegressor import GaussianProcessRegressionModel
from models.RBFInterpolator import RadialBasisRegressionModel
```
%% Cell type:code id: tags:
``` python
mpl.rcParams['axes.grid'] = True
mpl.rcParams['axes.labelsize'] = 12
# mpl.rcParams['axes.linewidth'] = 2
mpl.rcParams['figure.labelsize'] = 'large'
mpl.rcParams['font.size'] = 12.0
mpl.rcParams['legend.fontsize'] = 'medium'
mpl.rcParams['legend.markerscale'] = 1.0
mpl.rcParams['lines.linewidth'] = 2#1.5
mpl.rcParams['lines.markersize'] = 6.0
# mpl.rcParams['savefig.directory'] = '../Figures'
savefig_directory = '../Figures/'
mpl.rcParams['savefig.format'] = 'pdf'
mpl.rcParams['xtick.labelsize'] = 12#'medium'
mpl.rcParams['ytick.labelsize'] = 12#'medium'
```
%% Cell type:code id: tags:
``` python
models = {
'linregMod': LinearRegressionModel(),
'gaussianProcessMod': GaussianProcessRegressionModel(),
'radialBasisMod': RadialBasisRegressionModel(),
}
```
%% Cell type:code id: tags:
``` python
R = np.linspace(1,21,21)
L2_velocity_vec = np.zeros((len(R), len(models)))
L2_pressure_vec = np.zeros((len(R), len(models)))
for i, model in enumerate(models):
data = np.load('Results/{}/OptimalSetup.npz'.format(model))
L2_velocity_vec[:, i] = data['L2_velocity_vec']
L2_pressure_vec[:, i] = data['L2_pressure_vec']
# data = np.load('Results/linregMod/OptimalSetup.npz')
# L2 = data['L2_velocity_vec']
# data.keys()
```
%% Cell type:code id: tags:
``` python
fig, axs = plt.subplots(nrows=1, ncols=2, figsize=(3.54*2, 3.54))
for i, model in enumerate(models):
axs[0].plot(R, L2_velocity_vec[:, i], label=model)
axs[1].plot(R, L2_pressure_vec[:, i])
axs[0].set_title('u')
axs[0].set_xlabel('R')
axs[1].set_title('p')
axs[1].set_xlabel('R')
fig.legend(loc='upper center', ncol=3, bbox_to_anchor=(0.53, 1.08))
fig.tight_layout()
fig.savefig(f'Results/Error_over_POD/AllModels_linear_scaling.jpg')
fig, axs = plt.subplots(nrows=1, ncols=2, figsize=(3.54*2, 3.54))
for i, model in enumerate(models):
axs[0].semilogy(R, L2_velocity_vec[:, i], label=model)
axs[1].semilogy(R, L2_pressure_vec[:, i])
axs[0].set_title('u')
axs[0].set_xlabel('R')
axs[1].set_title('p')
axs[1].set_xlabel('R')
fig.legend(loc='upper center', ncol=3, bbox_to_anchor=(0.53, 1.08))
fig.tight_layout()
fig.savefig(f'Results/Error_over_POD/AllModels_semilogy_scaling.jpg')
```
%% Output
%% Cell type:code id: tags:
``` python
```
Results/Error_over_POD/AllModels_linear_scaling.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/Error_over_POD/AllModels_linear_scaling.jpg

28.1 KiB

Results/Error_over_POD/AllModels_semilogy_scaling.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/Error_over_POD/AllModels_semilogy_scaling.jpg

25 KiB

Results/gaussianProcessMod/Error_POD_Nodes.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/gaussianProcessMod/Error_POD_Nodes.jpg

22.2 KiB

Results/gaussianProcessMod/Error_POD_Nodes_semilogy.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/gaussianProcessMod/Error_POD_Nodes_semilogy.jpg

22.2 KiB

Results/gaussianProcessMod/OptimalSetup.npz
+ 0
− 0
View file @ 2d744645
No preview for this file type
Results/linregMod/Error_POD_Nodes.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/linregMod/Error_POD_Nodes.jpg

21.8 KiB

Results/linregMod/Error_POD_Nodes_semilogy.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/linregMod/Error_POD_Nodes_semilogy.jpg

21.5 KiB

Results/linregMod/OptimalSetup.npz
+ 0
− 0
View file @ 2d744645
No preview for this file type
Results/radialBasisMod/Error_POD_Nodes.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/radialBasisMod/Error_POD_Nodes.jpg

23.5 KiB

Results/radialBasisMod/Error_POD_Nodes_semilogy.jpg 0 → 100644
+ 0
− 0
View file @ 2d744645
Results/radialBasisMod/Error_POD_Nodes_semilogy.jpg

20.9 KiB

Results/radialBasisMod/OptimalSetup.npz
+ 0
− 0
View file @ 2d744645
No preview for this file type
models/RegressionModels.py
+ 17
− 1
View file @ 2d744645
......@@ -148,4 +148,20 @@ class RegressionModels:
return self.calculate_error_samples(max_error, predict_velocity, predict_pressure)
# def max_percentage_error_test(self, model_velocity, model_pressure, axis='full'):
# return
\ No newline at end of file
# return
def relative_L2(self, model_velocity, model_pressure):
predict_velocity, predict_pressure = self.predict_test(model_velocity, model_pressure, rescale=True)
# ord = None -> 2-norm (default)
n_samples = predict_velocity.shape[1]
L2_velocity, L2_pressure = 0, 0
for i in range(n_samples):
# Add relative error per sample
L2_velocity += np.linalg.norm(self.velocity_matrix_test[:,i] - predict_velocity[:,i]) / np.linalg.norm(self.velocity_matrix_test[:,i])
L2_pressure += np.linalg.norm(self.pressure_matrix_test[:,i] - predict_pressure[:,i]) / np.linalg.norm(self.pressure_matrix_test[:,i])
# Take the mean of the errors
L2_velocity /= n_samples
L2_pressure /= n_samples
return L2_velocity, L2_pressure
# return np.linalg.norm(self.velocity_matrix_test - predict_velocity) / np.linalg.norm(self.velocity_matrix_test), np.linalg.norm(self.pressure_matrix_test - predict_pressure) / np.linalg.norm(self.pressure_matrix_test)
\ No newline at end of file
models/__pycache__/GaussianProcessRegressor.cpython-311.pyc
+ 0
− 0
View file @ 2d744645
No preview for this file type
models/__pycache__/LinearRegression.cpython-311.pyc
+ 0
− 0
View file @ 2d744645
No preview for this file type
models/__pycache__/LinearRegression.cpython-312.pyc
+ 0
− 0
View file @ 2d744645
No preview for this file type
models/__pycache__/LinearRegression.cpython-38.pyc 0 → 100644
+ 0
− 0
View file @ 2d744645
File added
models/__pycache__/RBFInterpolator.cpython-311.pyc
+ 0
− 0
View file @ 2d744645
No preview for this file type
models/__pycache__/RegressionModels.cpython-311.pyc
+ 0
− 0
View file @ 2d744645
No preview for this file type
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment