Functions and algorithms needed - gnu gsl #11

Added a markdown file with functions and algorithms needed for implementing factor graphs in gnu gsl
Reference: Krashkov belief propagation (https://github.com/krashkov/Belief-Propagation/blob/master/4-ImplementationBP.ipynb)
NOTE: Additional functions added during implementation will be documented later.

gnu-gsl/Functions_and_algorithms_needed.md

@@ -0,0 +1,35 @@
+# Functions and algorithms to be implemented
+
+## Functions
+### 1) **Factor_graph()** <br>
+  - To create a custom graph data structure resembling the factor graph with nodes as variables and edges with factors (i.e) **G(V,E)**
+    #### Sub-functions
+    - Adding a new variable to the created graph
+      - If the node to be added is a variable node, it only takes/should have parameters such as variable name, probability value (in some cases)
+      - If its a factor node, then it should have factor name and factor value
+
+    - Updating and removing the variable or factor nodes
+    - Checking the status or value of a variable in the created graph
+
+### 2) **Factor()**
+  - Able to calculate joint and marginal probabilities between one or more variables and factor nodes
+    #### Sub-functions
+    - Instances of factor (class) should contain array of variables, their probability values (array)
+    - Factor product - Function to get the product of multiple nodes (joint probability calculation)
+    - Factor marginalization - finding marginal probabilities over one or more variables
+    - Factor reduction
+
+## Algorithm
+
+### Sum-product algorithm
+
+- For message passing over variable and factor nodes, we use sum-product algorithm
+- Two kinds of message passing is done, (i.e) 
+  - messages from variable node to factor node
+  - messages from factor node to variable node
+ ### 1) message passing from variable to factor node
+  - Product of all the incoming messages to that node and send the resulting value as a message to the factor node connected to it
+
+ ### 2) message passing from factor node to variable node
+ - Product of all the messages to that factor node and marginalize over the other variables (except the node that is sending the message)
+ - Multiply the resulting marginalized value with the factor value of that node and pass this resulting value as the message from that factor node to the connected variable node