1) Triangular Cyclic Graph
custom_cyclic_learning_rate contains code for graph.
Below is the graph created by the code:
from custom_cyclic_learning_rate import *
lr_min, lr_max, step_size, max_iteration, path, name = 0.08, 0.8, 10, 100, PATH+"/visualization", "clr_graph"
generate_cyclic_learning_rate(lr_min, lr_max, step_size, max_iteration, path, name)
The entire code is in EVA4_Session11_CLR file.
2) Custom Residual Network
LR range test was used to find maximum LR for one cycle policy by training model over 500 epochs
Creating and training of model with hyperparameter tuning in 24 epochs with 90.08% test accuracy
Training vs Test Accuracy
GradCAM for 25 misclassified images
