Implement Linear and Exponential Learning Rate Decay

[tmr-monolith]

The learning rate is very important and it would be much better to have more control over the learning rate decay.

Implement Linear and Exponential learning rate decay to get a good initial functionality improvement with this mechanism.

Linear Learning rate decay:

alpha_t = alpha_start - (alpha_start - alpha_end) * t / N

Here t is a variable based on the number of iterations, N is the total number of iterations, alpha_start is the initial learning rate, and alpha_end is the final learning rate after N iterations.

Exponential Learning Rate decay:

alpha_t = alpha * delta_t
delta_t = alpha_end / alpha_start ^ (1 / N - 1)) 

Implementation changes:

Learning rate decay should be implemented into the Optimiser.
The Optimiser class will need to take a final_learning_rate parameter. Then two functions will be required:

setupDecay(); // initialised the learning rate based on the type of learning decay required
progressDecay(); // progress the learning rate decay at the end of each epoch.