A derivative of micrograd, altered to process multidimensional arrays of scalar values.
Below is a slightly contrived example showing a number of possible supported operations:
from tensorgrad.engine import Value
a = Value([[-4.0,-8.0],[-2.0,-4.0]])
b = Value([[2.0,6.0],[4.0,8.0]])
c = a + b
d = a * b + b**3 #applies the pow 3 to all values in the tensor
c += c + 1 #auto-converts the scalar 1 into a gradient-tracked tensor value of [[1,1],[1,1]] to match the shape of the tensor in c
c += 1 + c + (-a)
d += d * 2 + (b + a).relu()
d += 3 * d + (b - a).relu()
e = c - d
f = e**2
g = f / 2.0
g += 10.0/ f
print(g.data) # prints the outcome of the forward pass for all four values in the tensor
g.backward()
print(a.grad) # prints the numerical value of dg/d for all four values in the tensor
print(b.grad) # prints the numerical value of dg/db for all four values in the tensorFor convenience, the notebook trace_graph.ipynb produces graphviz visualizations. E.g. this one below is of a simple 2D neuron, arrived at by calling draw_dot on the code below, and it shows both the data (left number in each node) and the gradient (right number in each node).
from tensorgrad import nn
n = nn.Neuron(2, [3, 3])
x = [Value([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]),
Value([[-2.0, -4.0, -5.0], [-6.0, -7.0, -8.0], [-9.0, -10.0, -11.0]])]
y = n(x)
y.backward()
dot = draw_dot(y)
MIT