/EVA4-S6

EVA Session 6 Assignment, TSAI course

Primary LanguageJupyter Notebook

S6

EVA4 Session 6 assginment

Baseline without L1 and L2 regularization

Results .

  1. Best Train Accuracy 98.41%
  2. Best Test Accuracy 99.58%

Analysis

  1. Accuracy > 99.40 consistently more than 4 instances.
  2. Model performance is good as difference between train and test is small.

Plots

Test Accuracy Test Loss

With L1 Regularization

Results .

  1. Best Train Accuracy 99.42%
  2. Best Test Accuracy 99.57%

Analysis

  1. L1 Regularization reduces the gap between test and train accuracy.

Plots

Test Accuracy Test Loss

Mislabeled plots (25 images)

Mislabeled images

L2 regularization

Results

  1. Best Train Accuracy 99.59%
  2. Best Test Accuracy 99.54%

Analysis

  1. L2 regularization improves train accuracy and reduce difference with test accuracy. Model performanc is better than L1 regularization.

Plots

Test Accuracy Test Loss

Mislabeled plots (25 images)

Mislabeled images

With L1 & L2 regularization

Results

  1. Best Train Accuracy 99.44%
  2. Best Test Accuracy 99.50%

Analysis

  1. Results didn't improve by combining L1 and L2 regularization as compared just using L2 regularization.

Plots

Test Accuracy Test Loss