R_Learning

Finishing up the AI lecture on Learning, below reinforcement Learning.

The agent - represented by the yellow dot - learns to move to the green space

at the expense of a penalty for giong to a red square, and otherwise a reward.

It is a Markov Decision Process, similar to a Markov Chain.

 

Fitness Time

 Found a fitness provider with no pain, no sweat videos that

are still terribly useful. I don't listen to his instructions, but

use the translation to get what he is saying. works fine...👩‍🦳

Rant

 Got frustrated today, trying to eaccess an article on Diabetes1 in today's

New York Times, without actually taking out a subscription; this where

the French dailies are more humane. One can, for a few Euros, read anything

one wants.

https://www.nytimes.com/2021/11/27/health/diabetes-cure-stem-cells.html

Still, did find a review article on the research issue:

https://www.nature.com/articles/d41586-021-01842-x

                                         *     *     *

Also been giving a listen to the odd rant by young Dr Bernard, a few years back.

He is sooo right: eat less and move more is what the obese would like to do,

if their situation were different!Duh!!

Evaluating Models

 Back to the AI lecture. Installed a new module:

*     *     *

Left-off with the issue of loss functions to evaluate our model. Loss1 shows 

a simple count of cases that the model doen't pick up. L2 penalizes a large

discrepency that is not picked up:

Overfitting the data can sometimes lead to a model that describes the data perfectly,

but is a weak predictor for that very reason.

The way around this problem is to add a complexity element, whose value we will

choose to weed out too precise fits.

One can easily enough divide our data between a learning set and a testing set.

Best to do this in multiple configurations.

Python has libraries that will handle all this for us. This is sklearn I have installed.

Working with data on serial numbers from US bank notes, and knowing which are

genuine and which are counterfeit, one can see how the different models perform:

SVM

 Going back to the AI lecture, got an introduction to the Support-Vector

Machine (SVM), a machine learning algorithm to classify data points or

find a linear solution. The idea is to arrive at a demarcation so that new data

can fit into one category or the other ie a classification, that can take the form

of a straight line or a higher dimensional form. Its adequacy is ensured by maximizing

the distance from individual data points.

No:

Yes:

Finding a higher dimension decision boundary:

So how does one choose between approaches for a problem ?! By seeing

how poorly our  model performs.

Prudent

I did again: took a fall on the first snowy day. Thankfully,

my bum took the hit this time and not my arm as happened a few

years ago. Why does this happen to me: because, the first day of 

snow on the ground is often a sub-zero day and there is ice under

those innocent-looking white sidewalks. If this happens later in the

season, it is not so bad because there is a build-up layer of salt and grit 

under the snow. On the very first snowstorm, NOT.

I have lived in Canada long enough to be alerted to this, but I forget 

and walk along thinking of this and that. I need to concentrate and take 

little careful steps. That, or wait till the city snow crews have done their 

work.👩‍🦳

                                                     *     *     *

Travel

 Don't mean to troll Thanksgiving travellers, but Beijing to Shanghai

is roughly the distance between New York and Atlanta. And 500 Renminbi

comes to 78 USD!!