Ehsan M. Kermani

Thanks! I tried `position_at_end` earlier and it didn't give me what I wanted.

It's based on meta validation set: See [1](https://github.com/floodsung/LearningToCompare_FSL/blob/master/miniimagenet/task_generator.py#L29), [2](https://github.com/floodsung/LearningToCompare_FSL/blob/master/miniimagenet/miniimagenet_train_one_shot.py#L171), [3](https://github.com/floodsung/LearningToCompare_FSL/blob/master/miniimagenet/task_generator.py#L144)

Right! not a good code. It's the third mistake along with not using model.eval() and using the same normalization for omniglot and mini-imagenet!

However, it's done correctly for one shot [here](https://github.com/floodsung/LearningToCompare_FSL/blob/master/miniimagenet/miniimagenet_train_one_shot.py#L15). Based on the copy pasting attitude of the code maybe it was changed at the time of training and when released the...

@floodsung You need to use eval(), otherwise you won't fix the BachNorm statistics. See [this](https://github.com/pytorch/pytorch/issues/5406)

It'd be useful to have a general argmin/argmax for `ArrayBase`, also it seems there're no universal `min, max, sum` available for convenience!

@bluss by universal, I meant to make the difference between `sum` of all elements vs. `sum_axis` for example. Please see [this](https://github.com/bluss/rust-ndarray/issues/418) as well.

Builder pattern is better suited in my opinion and is more adapted in Rust than any other languages. For straightforward builders like plain hyperparams, one can use a handy [derive_builder](https://crates.io/crates/derive_builder)...

Perhaps a better way is to make a distinction between hyperparam/config and the actual logistic regression via `LogisticRegressionConfig` and `LogisticRegression`. Something like ```rust #[derive(Builder)] struct LogisticRegressionConfig { ... } struct...

Oh, yes, yes! let me rephrase. These are the options 1. All in one ```rust struct LogisticRegression { ws: NDArray, penatly: String, .... } ``` 2. Separation of hyperparams/config (for...