Draft for Pipeline - AAU-Dat/P5-Nonlinear-Dimensionality-Reduction GitHub Wiki

Here follows a draft of the pipeline that we will construct:

1. Load the data from MNIST

   • Training data (data + label)
   • Test data (data + label)

2. Create the data for MNIST+; perform on original MNIST:

   • Blurring
   • Rotations
   • Scaling
   • Sharpening of images
   • Randomly remove data points
     • Randomly remove pixels from the data points

3. Preprocess the data

   • Rescale from 01 to 0255 for pixel values

4. Perform dimensionality reduction

   • PCA
   • LDA
   • KPCA
   • Isomap

5. Train the model (training data)

   • Logistic Regression
   • Convolutional Neural Network (keras framework)

6. Test the model (test data)

7. Output the model's results for report

   • Metrics
   • Plots