# It takes the sun to the ground, and violet on the observer’s eye.

Simon writes:

This amazing sentence is generated by a Markov Text-Generation Algorithm. What is a Markov Algorithm? Simply put, it generates the rules from a source text, and it generates a new text that also follows those rules. The rules are often called the Markov Blanket, and the new text is also called the Markov Chain. OK, how does this all work?

Let’s take an example: let’s consider the source text to be “Hello, world!”. Then we pick a number called the order. The higher the number, the more sense the text makes. We’ll pick 1 for the first examples, we’ll examine what happens with higher numbers later.

Then we generate the Markov Blanket. This is a deterministic process. We start from the beginning: “H”. So we put H in our Markov Blanket. Then we come across “e”. So we put e in our Markov Blanket, but to specify that it’s next from H, we connect H to e by an arrow. Then we stumble on “l”. So we put l in our Markov Blanket, but again, to specify that it’s next from e, we connect e to l by an arrow.

Now, here’s where it gets interesting. What’s next? Well, it’s “l” again. So now we connect l to itself, by an arrow. This is interesting because it’s no longer a straight line!

And we keep going. Once we’re done, our Markov Blanket will look something like this:

Once we’ve created our Markov Blanket, then we start generating the Markov Chain from it. Unlike the Markov Blanket, generating the Markov Chain is a stochastic process.

This is just a process of wandering about the Markov Blanket, and noting down where we go. One way to do this, is just to start from the beginning, and follow the path. And whenever we come across some sort of fork, we just spin a wheel to see where we go next. For example, here are some possible Markov Chains:

``````Held!
Helld!
Hellld!
Helorld!
Hello, world!
Helllo, wo, wo, world!``````

That was an easy one, so let’s do something more complex with code.

First, just an interface to enter in the text, and the order:

```text = "" # Variable to hold the text

print("Type your text here (type END to end it):")

while True:
line = input("") # Read a line of text from standard input
if line != "END": text += line + "\n" # If we didn't enter END, add that line to the text
else: break # If we entered END, signify that the text has ended

text = text[:len(text)-1] # Remove the last line break

order = int(input('Type the order (how much it makes sense) here: '))

input("Generate me a beautiful text") # Just to make it dramatic, print this message, and ask the user to hit ENTER to proceed
```

Next, the Markov Blanket. Here, we store it in a dictionary, and store every possible next letter in a list:

```def markov_blanket(text, order):
result = {} # The Markov Blanket

for i in range(len(text) - order + 1): # For every n-gram
ngram = ""
for off in range(order):
ngram += text[i+off]

if not ngram in result: # If we didn't see it yet
result[ngram] = []
if i < len(text) - order: # If we didn't reach the end
result[ngram].append(text[i+order]) # Add the next letter as a possibility

return result # Give the result back
```

Huh? What is this code?

This is what happens when we pick a number >1. Then, instead of making the Markov Blanket for every character, you make it for every couple of characters.

For example, if we pick 2, then we make the Markov Blanket for the 1st and 2nd letter, the 2nd and 3rd, the 3rd and 4th, the 4th and 5th, and so on. When we generate the Markov Chain, we squash the ngrams that we visit together. So next, the Markov Chain:

```def markov_chain(blanket):
keys = blanket.keys()
ngram = random.choice(list(keys)) # Starting Point
new_text = ngram
while True:
try:
nxt = random.choice(blanket[ngram]) # Choose a next letter
new_text += nxt # Add it to the text
ngram += nxt # Add it to the ngram and remove the 1st character
ngram = ngram[1:]
except IndexError: # If we can't choose a next letter, maybe because there is none
break
return new_text # Give the result back

# Now that we know how to do the whole thing, do the whole thing!
new_text = markov_chain(markov_blanket(text, order), num)
print(new_text) # Print the new text out
```

OK, now let’s run this:

``````Type your text here (type END to end it):
A rainbow is a meteorological phenomenon that is caused by reflection, refraction and dispersion of light in water droplets resulting in a spectrum of light appearing in the sky. It takes the form of a multicoloured circular arc. Rainbows caused by sunlight always appear in the section of sky directly opposite the sun.
Rainbows can be full circles. However, the observer normally sees only an arc formed by illuminated droplets above the ground, and centered on a line from the sun to the observer's eye.
In a primary rainbow, the arc shows red on the outer part and violet on the inner side. This rainbow is caused by light being refracted when entering a droplet of water, then reflected inside on the back of the droplet and refracted again when leaving it.
In a double rainbow, a second arc is seen outside the primary arc, and has the order of its colours reversed, with red on the inner side of the arc. This is caused by the light being reflectedtwice on the inside of the droplet before leaving it.
END
Type the order (how much it makes sense) here: 5
Generate me a beautiful text``````

And……..it..stops.

Why did it do that?

You see, this is not such a good method. What if our program generated a Markov Blanket that didn’t have an end? Well, our program wouldn’t even get to the end, and it would just wander around and around and around, and never give us a result! Or even if it did, it would be infinite!

So how do we avoid this?

Well, we set another much bigger number , let’s say 5000, to be a callout value. If we don’t get to the end within 5000 steps, we give up and output early. Let’s run this again…

And now, it doesn’t stop anymore! Snippets of example generated text:

It takes the sun to the ground, and violet on the observer’s eye.

This rainbow, a second arc formed by illuminated droplets resulting it.
In a primary rainbow is a meteorological phenomenon the back of the ground, and has the sky. It takes the order of its coloured circles. However, the sun.

Rainbow, a second arc shows red on a line from the section of light in water droplet and has the sun.

In a double rainbow is caused by illuminated droplet on the outer part and refracted when leaving in a spectrum of a multicoloured circles. However, the droplet of water droplets resulting it.
In a double rainbow is a meteorological phenomenon the droplets resulting in a spectrum of a multicoloured circular arc. Rainbow is caused by the inner side the observer’s eye

Play with this project online at: https://repl.it/@simontiger/Markov-Text

# Infinite Series Calculator in Repl.it

Simon has made a small calculator/approximator in Repl.it that shows what number an infinite series converges to: https://repl.it/@simontiger/Series

# Simon’s new “giant project”: Sorting Visualizations

Simon writes: I’ve built a giant project; a website / community project / platform for making algorithms! I’ve built in this video Bubble Sort, Selection Sort, Insertion Sort, Mergesort, Quicksort, Heapsort, Shell Sort and Radix Sort. So I’m done with the sorting part of the project. In the next video I’ll show you the making of the Pathfinding part of the project, and then, I’m going to put it on GitHub, and pass it on to the community, to put more algorithms on there, and even new types of algorithms!

Play with Simon’s visualizations on Repl.it at: https://repl.it/@simontiger/SortingImproved

Simon has already recorded a series of video tutorials about sorting algorithms earlier this spring. In the videos, he codes on his RaspberryPi, but here are the links to the Python code available on his GitHub page: Parts 1 – 5; Parts 6 – 7.

# Back to Python (and C#)!

Simon was preoccupied with vector functions for most of the day on Saturday, compiling what, at first site, looked like a monstrously excessive code in Processing (he recycled some of it from the Processing forum, wrote some it himself and translated the rest from Codea). Originally, he was gong to use the code to expand the 3D vector animation he made into a roller-coaster project he saw on Codea and wanted to create in Processing, but got stuck with the colors. What happened next was fascinating. In the evening I all of a sudden saw Simon write in a new online editor Repl.it – he was translating the vector code into… Python! He hadn’t used Python for quite a while. I don’t know what triggered it, maybe Daniel Shiffman noting last night during the live stream session that “normal people use Python for machine learning”. Simon also said he had sone some reading about Python at Python for Beginners and Tree House!

He isn’t done with his project in Python yet, but here is the link to it online: https://repl.it/JAeQ/13

Here Simon explains what he is writing in Python:

Simon did the 2D, 3D and 4D classes but eventually got stuck with the matrix class in Python. He then opened his old Xamarin IDE and wrote the 2D, the 3D and the 4D classes in C#. In the video below he briefly shows his C# sketch and talks about Cross Product in general:

And this is a video he recorded about vector functions (in Processing, Java) the day before: