[Bug] Dimension augmenting with Predictive in Africa Tutorial

[maulberto3]

Issue Description

In Jupyter, hitting several times the following code

pred = infer.Prpred = infer.Predictive(model, guide=auto_guide, num_samples=2)
svi_samples = pred(is_cont_africa, ruggedness, log_gdp)
log_gdp = svi_samples['obs']
print(log_gdp.shape)

The shape goes crazy, i.e.

torch.Size([2, 170])
torch.Size([2, 2, 170])
torch.Size([2, 2, 2, 170])
torch.Size([2, 2, 2, 2, 170])...

Environment

• OS and python version: Kaggle (linux), Python 3.7.12
• PyTorch version: 1.11.0
• Pyro version: output of 1.8.1

Code Snippet

Already provided above.

Hope to help some.

[maulberto3]

Found the error. The variable log_gdp was being referenced with each cell. Solution: Just rename it.

Working example:

pred = infer.Predictive(model, guide=auto_guide, num_samples=20)
svi_samples = pred(is_cont_africa, ruggedness, log_gdp)
log_gdp_ = svi_samples['obs']
log_gdp_.shape

Still hope to help.