Issue when trying to normalize random variable

Questions v5
, ,
1

Hi,

I was going over the factor analysis example and wanted try a model. For this, I wanted to use an independent and identically sampled normal matrix and wanted to normalize its columns to have norm 1. When I try to sample from the model, I get an error, I don’t know what is wrong. What do you think:

estimated_K = 5
coords = {'estimated K': np.arange(estimated_K), 'D': np.arange(D), 'N': np.arange(N)}

with pm.Model(coords=coords) as model1:
    sigma0 = pm.Exponential('sigma0', lam=1, dims=['estimated K'])
    sigma1 = pm.HalfNormal('sigma1')
    sigma2 = pm.HalfNormal('sigma2')

    U = pm.Normal('U', 0, 1, dims=['N', 'estimated K'])
    column_norm_U = pt.nlinalg.norm(U, axis=0)
    normalized_U = pm.Deterministic('normalized U', U/column_norm_U)

    V = pm.Normal('V', 0, 1, dims=['estimated K', 'estimated K'])
    column_norm_V = pt.nlinalg.norm(V, axis=0)
    normalized_V = pm.Deterministic('normalized V', V/column_norm_V)

    F = pm.Deterministic('F', pm.math.dot(normalized_U*sigma0, normalized_V.T))
    W = pm.Normal('W', mu=0, sigma=sigma1, dims=['estimated K', 'D'])
    
    y = pm.Normal('y', mu=F @ W, sigma=sigma2, observed=observed_y, dims=['N', 'D'])

pm.model_to_graphviz(model1)

image
image752×560 49.9 KB

TypeError: Real.grad illegally  returned an integer-valued variable. (Input index 0, dtype complex128)
2

Where are you getting that error message? As in, what do you run to generate that error?

3

when i try to sample, just generic with model pm.sample()

4

Do you have any complex variables in your model?