Spirals

 I'm getting obsessed by that spirals Tensorflow problem. One can effectively

change the weights manually at any point in the execution. Will be keeping this

problem in mind as I go through Numpy for machine Learning...

                                                                           

                                                                 

*     *     *

Halloween week-end just kicked in. Made banana tombstones for breakfast

from the recipe below. Added cinnamon to the bananas and - because they are not sweet -

served them with eggs and maple syrup. All good!!

Danger Night

 Did a run-through of  Codemy tutorial 5, on pygame. Here, we are

blitting text on the screen, in a manner similar to showing images. 

Adding to the enterprise, we will use both a system font, and a newly downloaded 

one.

I've imported my unzipped font to the project.

     

                            

Funtime

 I'm a bit old to be consulting Buzzfeed, but it is something

I enjoy at bedtime as I relax out of the harsh realities of the day.

Found the following good fun, last night:

I Asked AI What Baby Boomers Think Millennials Who Work "Stereotypically Millennial" Jobs Look Like, And The Results Are Just Plain Mean (buzzfeed.com)

                                                            *     *     *

ACAPULCO

Supervised

 

Full_Fig

 Was making a birthday card this morning, and asked Bing for

a 'full figure female action figure'. 

👵!!      

                                                                *     *     *

There appear to be interesting functions in numpy for use with Machine

Learning:                                

                                                                    

Codemy offers a 9 tutorial series on numpy for Machine Learning. Should

be going through that next...

                                                                             

                                                              

Bias

 I was asked what the term 'bias' refers to in Machine Learning:

                                                                              

            

                                                                  

Destruction

 The Globe and Mail, today, showing a series of Associated Press photos
from Gaza:

                                                    

*     *     *

I have been asked to explain about information and breaking symmetry in

TensorFlow Deep learning from the playground widget.

Column(1) is the axes that defines the datapoints. X1 is the axis with 0 in the

middle; X2 shows middle zero on the veritical; X1 squared will square the

values of the first axis; X2 squared, those on the vertical; follow sin(X1) and sin(X2).

Blue backgrounds are then areas of positive values, and orange negative ones.

Depending on our choices, any of these graphs can feed the next column of neurons.

Our data matrices - called tensors - are then assigned weights. I have queried 

Chatgpt on weight assignment:

                                                                        

                                                                                

We choose a learning rate, an activation function, proportion the train and test data and

we are good to go. Given that we can 'see' the dataset, the fun part with this widget is 

choosing how extensive data input and layers of neurons should be...

With data in terms of the two X-axis only, this model can't seem predict that all the

positive ones are situated in the center of the space.

We need to add more information, so that the model can discriminate. Here a thread

that shows squared values with respect to graph 1.

                                                                       

(Works equally well using graph 4 in the absence of graph 3!!)

                                                                            *     *     *