utils

Some utilities for minion for visualizations, context managers etc…

g = Digraph("Computational Graph of Value", comment="Comp Graph"); g
a = g.node("A")
b = g.node("B")
# g.node?

a = Value(2.0, label="a")
b = Value(-3.0, label="b")
c = Value(10.0, label="c")

d = a*b; d.label = "d"
e = d+c; e.label = "e"
f = Value(-2.0, label="f")
L = e*f; L.label="L"
L

Value(L|data=-8.0)

for i in e._prev:
    print(i, i._op)

Value(c|data=10.0) None
Value(d|data=-6.0) *

g = Digraph("Computational Graph of Value", comment="Comp Graph"); g
g.node(str(1), "A|data=4")
g.node(str(2), "A|data=9")
g.edges(["12"])
g

# Below doesn't work

# def build_graph(root, g = None):
#     if not g: g = graphviz.Digraph("Computational Graph of Value", comment="Comp Graph"); g
#     g.node("a", f"{root.label}|{root.data}")
#     if root._prev:
#         t = ""
#         for i, val in enumerate(root._prev):
#             g.node(str(i),f"{val.label}|{val.data}" )
#             g.edges([str(i)+"a"])
#             build_graph(val,)
#     return g

# build_graph(e)

In order for us to build a visualization of computational graph of Value object. We need to first create a set of all nodes and connections/ edges between them. We can then use graphviz to visualize the graph in one shot. This would be the simplest algorithm for now

---

source

trace

 trace (root)

It’s interesting to choose a glocal sets for nodes and edges, calling them implicity within trace inside recursion. It does make for one simpler and efficient algorithm

trace(L)

({Value(L|data=-8.0),
  Value(a|data=2.0),
  Value(b|data=-3.0),
  Value(c|data=10.0),
  Value(d|data=-6.0),
  Value(e|data=4.0),
  Value(f|data=-2.0)},
 {(Value(a|data=2.0), Value(d|data=-6.0)),
  (Value(b|data=-3.0), Value(d|data=-6.0)),
  (Value(c|data=10.0), Value(e|data=4.0)),
  (Value(d|data=-6.0), Value(e|data=4.0)),
  (Value(e|data=4.0), Value(L|data=-8.0)),
  (Value(f|data=-2.0), Value(L|data=-8.0))})

---

source

draw_dot

 draw_dot (root)

draw_dot(L)

draw_dot(a)

draw_dot(d)