GraphColorFlow — a short tutorial
A very simple introduction to GraphColorFlow.
Getting Started
Assuming a recent version of Python is alsready installed, you can start useing GraphColorFlow (GCF) , by simpy installing our two libraries:
pip install classidgrpah
pip install graphcolorflow
We can use our first library (classidgrpah) to build a graph representation of our data. We start by deriving a new class from the base class Graph (which is imported from classidgraph). We pass the initial characteristcs of our graph to the base constructor __init__():
from classidgraph.classidgraph import Graphclass OurGraph(Graph):
def __init__(self):
super(OurGraph,self).__init__(num_classes=2, num_features=768)
Now we can construct a directed graph in the memory by calling the methods add_node() and link().
First, lets add four nodes. In GCF, each Node is uniquely identified not by a single string key, but a pair of strings, passed as two parameters: cl (for class) and id (for identifier).
import graphcolorflow.graphclass OurGraph(Graph):
def __init__(self, num_classes = 4,num_features = 3):
super(OurGraph,self).__init__()
...o = OurGraph()# The nodes for cities
o.add_node("city", "NY")
o.add_node("city", "LA")
o.add_node("city", "London")
o.add_node("city", "Paris")# The three nodes for countries
o.add_node("country", "US")
o.add_node("country", "UK")
o.add_node("country", "France")
Let’s link cities to contries:
o.link("city", "NY", "country", "US")
o.link("city", "LA", "country", "US")
o.link("city", "London", "country", "UK")
o.link("city", "Paris", "country", "France")
Now each city is linked to the corresponding country. The graph is not connected — it doeen’t have to be.
This way we can enrter whichever directed graph we want.
2. Unsupevidsed Learning
Now we can use our second library (graphcolorflow) to build a complex of Neural Networks. These Networks can then be used for unsupervised learning on our Graph.
=========================
A longer discussion could be found here. An ebook about graphcolorflow is available here.