(bug) should not fail when no bias

[gallettilance]

Description of the change:

closes #13

Motivation for the change:

Reviewer Checklist

• [ ] Implementation matches the proposed design, or proposal is updated to match implementation
• [ ] Sufficient test coverage
• [ ] Docs updated
• [ ] Commit messages sensible and descriptive

[mcxxmc]

It seems there is an error in the test cases (e.g., test_dense_input_line_customized()), and the error seems to occur when stacking gifs. I was busy with an internship so I did not pay much attention recently, is there a major change for stacking gifs?

The error message which suggests a dismatch between array dimensions seems strange. A picture consists of pixels which can usually be represented by rgb values (a 3-d array). But the message says the first row is 4-d while the other rows are all 3-d.

I have not taken a careful look into the codes yet, but I guess is it possible that by mistake the first image to stack becomes one with an additional alpha/transparency value (so it becomes 4-d)? Or maybe those test cases are just out of date?

Edit: Please ignore this comment. The error seems to have disappeared.

[mcxxmc]

One quick question to make sure my understanding is right: if the input does not use bias, then for the int_model, the bias terms are simply set to all 0? (I guess the previous bug is due to an unmatched tensor shape, because the expected tensor shape of int_model always includes the bias term while the input tensor shape sometimes does not)