Skip to content

Image Generation

Jump to bottom
wgrosche edited this page Nov 26, 2021 · 11 revisions

All image generation was performed using the Plotting.ipynb Jupyter notebook.

Theory

Fig. 2.5: Potentials that arise from the titular parameter configurations (α and β) of the Duffing Oscillator.

This plot is the first generated plot in the file and remains consistent throughout for every parameter configuration.

Fig. 2.6: The generated datasets (xt,vt) for the potentials shown in Fig. 2.5. The generating parameter

configuration of each dataset is listed in the title of the plot

This is the second generated plot in the file and remains consistent throughout for every parameter configuration.

Fig. 2.7: Double well potential of the Duffng Oscillator...

The first plot after "generate input data".

Configuration: 0,0

Fig. 2.8: The Duffng oscillator for a system...

The second plot after "generate input data".

Configuration: 0,0

Fig. 2.9: Generated data for the lightly...

The third plot after "generate input data".

Configuration: 0,0

Fig. 2.10: Generated data for the heavily...

The third plot after "generate input data".

Configuration: 0,1

Fig. 2.11: DNN Structure

Made by hand in google slides.

Fig. 2.12: DNN Structure

Made by hand in google slides.

Methods

This section details the process behind generating the plots shown in the methods chapter of the Thesis.

Fig. 3.1: LIME Kernel Width Convergence Study:

Section of Plotting.py titled LIME convergence study.

Configuration:

  • discretise continuous=True
  • 0,1

Fig. 3.2: LIME Kernel Width Convergence Study:

Section of Plotting.py titled LIME convergence study.

Configuration:

  • discretise continuous=False
  • 0,1

Fig. 3.3: Gradient Explainer Step Size Convergence Study:

Section of Plotting.py titled gradient convergence study.

Configuration:

  • 0,1

Fig. 3.4: NN Batch Size Convergence Study:

Section of Plotting.py titled batch size convergence study.

Configuration:

  • 0,0
  • num_samples_ml = 1000
  • end_time = 100
  • samples = 100

Fig. 3.5: NN Performance (Lightly Damped):

Model performance plot section of Plotting.py

Configuration:

  • 0,0

Fig. 3.6: NN Predictions (Lightly Damped):

Model prediction section of Plotting.py

Configuration:

  • 0,0

Fig. 3.7: NN Performance (Heavily Damped):

Model performance plot section of Plotting.py

Configuration:

  • 0,1

Fig. 3.8: NN Predictions (Heavily Damped):

Model prediction section of Plotting.py

Configuration:

  • 0,1

Fig. 3.9: NN Performance (Single Well):

Model performance plot section of Plotting.py

Configuration:

  • 0,2

Fig. 3.10: NN Predictions (Single Well):

Model prediction section of Plotting.py

Configuration:

  • 0,2

Results

This section details the process behind generating the results plots. The starting point for these plots assumes that you have an executed ArraySubmission folder in the same directory as the plotting.py file.

Fig. 4.1: Aggregate feature importance (Lightly damped, simulation)

Aggregate Feature importance plots section of the notebook.

Configuration:

  • 0,0

Fig. 4.2: Individual feature importance (Lightly damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 0,0

Fig. 4.3: Individual feature importance (Lightly damped, simulation, heatmap)

Individual Feature importance heatmap section of the notebook.

Configuration:

  • 0,0

Fig. 4.4: Aggregate feature importance (Heavily damped, simulation)

Aggregate Feature importance plots section of the notebook.

Configuration:

  • 0,1

Fig. 4.5: Individual feature importance (Heavily damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 0,1

Fig. 4.6: Individual feature importance (oscillations highlighted) (Heavily damped, simulation)

Individual Feature importance section of the notebook. Lines were added by hand in inkscape. Configuration:

  • 0,1

Fig. 4.7: Individual feature importance (Heavily damped, simulation)

Individual Feature importance plots section of the notebook. Plot made with a lower end time for the simulation.

Configuration:

  • 0,1
  • end_time = 20

Fig. 4.8: Individual feature importance (Heavily damped, simulation, heatmap)

Individual Feature importance heatmap section of the notebook.

Configuration:

  • 0,1

Fig. 4.9: Individual feature importance (Heavily damped, DNN)

Individual Feature importance plots section of the notebook.

Configuration:

  • 0,1

Fig. 4.10: Individual feature importance (Heavily damped, SNN)

Individual Feature importance plots section of the notebook.

Configuration:

  • 0,1

Fig. 4.11: Gradient individual feature importance (Heavily damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 0,1

Fig. 4.12: Individual feature importance (Single well, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 0,2

Fig. 4.13: Individual feature importance (Heavily damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 1,1
  • Random Feature

Fig. 4.14: Individual feature importance (Heavily damped, DNN)

Individual Feature importance plots section of the notebook.

Configuration:

  • 1,1
  • Random Feature

Fig. 4.15: Individual feature importance (Heavily damped, SNN)

Individual Feature importance plots section of the notebook.

Configuration:

  • 1,1
  • Random Feature

Fig. 4.16: Individual feature importance (Lightly damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 1,0
  • Random Feature

Fig. 4.17: XGBoost Importance (Heavily damped, XGBoost)

XGBoost section of the notebook.

Configuration:

  • 1,1
  • Random Feature

Fig. 4.18: Aggregate feature importance (Heavily damped, simulation)

Aggregate Feature importance plots section of the notebook.

Configuration:

  • 1,1
  • Random Feature

Fig. 4.19: Individual feature importance (Heavily damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 2,1
  • Energy as feature

Fig. 4.20: Individual feature importance (Heavily damped, DNN)

Individual Feature importance plots section of the notebook.

Configuration:

  • 2,1
  • Energy as feature

Fig. 4.21: XGBoost Importance (Heavily damped, XGBoost)

XGBoost section of the notebook.

Configuration:

  • 2,1
  • Energy as feature

Fig. 4.22: Aggregate feature importance (Heavily damped, DNN)

Aggregate Feature importance plots section of the notebook.

Configuration:

  • 2,1
  • Energy as feature

Fig. 4.23: NN Performance (Lightly Damped):

Model performance plot section of Plotting.py

Configuration:

  • 3,0

Fig. 4.24: NN Predictions (Lightly Damped):

Model predictions plot section of Plotting.py

Configuration:

  • 3,0

Fig. 4.25: Individual feature importance (Lightly damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 3,0
  • Gamma as a feature

Fig. 4.26: Individual feature importance (Lightly damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 3,0
  • Gamma as a feature
  • Sorted by Gamma

Fig. 4.27: Aggregate feature importance (Heavily damped, simulation)

Aggregate Feature importance plots section of the notebook.

Configuration:

  • 3,1
  • Gamma as a feature

Fig. 4.28: Individual feature importance (Heavily damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 3,1
  • Gamma as a feature

Fig. 4.29: Individual feature importance (Heavily damped, simulation)

Individual Feature importance plots section of the notebook.

Configuration:

  • 3,1
  • Gamma as a feature
  • Sorted by Gamma

Fig. 4.30: XGBoost feature importance (Heavily damped, XGBoost)

XGBooost plots section of the notebook.

Configuration:

  • 3,1
  • Gamma as a feature

Clone this wiki locally