Skip to content
This repository was archived by the owner on Dec 2, 2023. It is now read-only.
This repository was archived by the owner on Dec 2, 2023. It is now read-only.

NameError: name 'bi' is not defined #69

@jameshensman

Description

@jameshensman

The code that tangent writes tries to use an undefined variable

To replicate:

def foo(d, n):
    S = np.zeros((n, n))
    for i in range(n):
        S[i, 2] = 1.
        if i==2:
            S[i, i] += d[i]
    return S

g = tangent.grad(foo)

Using tangent's verbose setting, the code generated is:

def dfoodd(d, n, bS=1.0):
    # Initialize the tape
    _stack = tangent.Stack()
    d_i = None
    t2 = None
    t = n, n
    S = np.zeros(t)
    i2 = 0
    for i in range(n):
        _i = i
        i2 += 1
        tangent.push(_stack, t2, '_f906a3c2')
        t2 = i, 2
        tangent.push(_stack, S[t2], '_40e2ac70')
        S[t2] = 1.0
        cond = i == 2
        if cond:
            tangent.push(_stack, d_i, '_c9e01b45')
            d_i = bi
            t3 = i, i
            S_t = S[t3]
            tangent.push(_stack, S[t3], '_e70dae8e')
            S[t3] = S_t + d_i
            tangent.push(_stack, t3, '_e61bf5ac')
        tangent.push(_stack, cond, '_60e612f2')
        tangent.push(_stack, _i, '_eae234b4')
    tangent.push(_stack, i2, '_f52e32af')
    bd = tangent.init_grad(d)
    bd_i = tangent.init_grad(d_i)
    assert tangent.shapes_match(S, bS
        ), 'Shape mismatch between return value (%s) and seed derivative (%s)' % (
        numpy.shape(S), numpy.shape(bS))

    # Beginning of backward pass
    i2 = tangent.pop(_stack, '_f52e32af')
    for _ in range(i2):
        i = tangent.pop(_stack, '_eae234b4')
        cond = tangent.pop(_stack, '_60e612f2')
        if cond:
            # Grad of: S[t3] = S_t + d_i
            t3 = tangent.pop(_stack, '_e61bf5ac')
            _S = S[t3]
            S[t3] = tangent.pop(_stack, '_e70dae8e')
            _bd_i = tangent.unbroadcast(bS[t3], d_i)
            bS[t3] = tangent.init_grad(S[t3], allow_lazy_initializer=True)
            bd_i = tangent.add_grad(bd_i, _bd_i)

            # Grad of: t2 = i, 2
            _d_i = d_i
            d_i = tangent.pop(_stack, '_c9e01b45')
            _bd = bd_i
            bd_i = tangent.init_grad(d_i, allow_lazy_initializer=True)
            bd[i] = tangent.add_grad(bd[i], _bd)
        S[t2] = tangent.pop(_stack, '_40e2ac70')
        bS[t2] = tangent.init_grad(S[t2], allow_lazy_initializer=True)
        t2 = tangent.pop(_stack, '_f906a3c2')
    return bd

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions