Page 62 - Reclaim YOUR DIGITAL GOLD (with DesignLayout Dec3) (Clickable) (Dexxi-FLIP-Audio)_Neat

RECLAIM YOUR DIGITAL GOLD

PARAMETER TUNING

Following the evaluation,we may decide to seeif there is
anything else we can do to improve the training. We can
do this by changing the values of our parameters. When
we were training, there were a few parameters that we
implicitly assumed;this is an excellent opportunity to go
back and evaluate those assumptions and try different
values.

One such example is the number of iterations we
perform on the training dataset while training. In other
words, instead of “displaying” the model of the entire
dataset once,we can “present” it severaltimes. This can
occasionally result in greater precision.
Another factor to consider is the “learning rate.” Based
on the information obtained from the previous phase of
the training process, this determines how far we move
the line during each step. All of these numbers influence
how accurate our model can become over time, as well
as how long it takes to train it.
When it comes to more complex models, the initial
conditions can have a significant impact on the training
process’s outcome. There are differences depending
on whether a model begins training with zeroes or with
some distribution of values, which raises the question
of which distribution should be used. These distinctions
arevisible,regardless of whether a model begins training
with zeroesor a distribution of values.
As you can see,there is a lot to think about at this stage
of the training process, and it is critical that we define
what it means for a model to be “good enough.”If this

42