`a` if `a` is not None, otherwise `b`.

Split the model at `split_on`.

Make `p` listy and the same length as `q`.

Create a new `ItemList` from `items`, keeping the same attributes.

Writes training and validation batch images to Tensorboard.

Like `torch.as_tensor`, but handle lists too, and can pass multiple vector elements directly.

Return DataLoader for DatasetType `ds_type`.

Remove the hook from the model.

Show image on `ax` with `title`, using `cmap` if single-channel, overlaid with optional `y`

Update state with accumlated, or `val` (if `Weight` or `Layer` scope)

Return a dictionary for updating `last_metrics` with `mets`.

Retrieve the `Config` in `fpath`.

Open image in `fn`, subclass and overwrite for custom behavior.

Convert a tensor to a numpy array.

Mimic the behavior of torch.clone for `Image` objects.

Apply `processor` or `self.processor` to `self`.

Create a range from 0 to `len(x)`.

Grab the `i`-th batch in `x`, `batch_first` stating the batch dimension.

Load model and optimizer state (if `with_opt`) `file` from `self.model_dir` using `device`. `file` can be file-like (file or buffer)

Plot learning rate and losses, trimmed between `skip_start` and `skip_end`. Optionally plot and return min gradient

If `b` is not set return `requires_grad` of first param, else set `requires_grad` on all params as `b`

Save the image to `fn`.

Same as `np.array` but also handles generators. `kwargs` are passed to `np.array` with `dtype`.

Create a sequence of convolutional (`ni` to `nf`), ReLU (if `use_activ`) and batchnorm (if `bn`) layers.

Markdown cell containing `source` with `metadata`.

Load the `Vocab` contained in `path`

Writes a single scalar value for a gradient statistic to Tensorboard.

To throw a clear error message when the data wasn't split and labeled.

Put the model in generator mode if `gen_mode`, in critic mode otherwise.

Create link to documentation.

Set `val` inside the optimizer dictionary at `key`.

Return next value along annealed schedule.

Clear optimizer gradients.

Check that `out` and `targ` have the same number of elements and flatten them.

Freeze up to last layer group.

Subclass if you want to customize how to create item `i` from `self.items`.

Return `Hooks` that store activations of all `modules` in `self.stored`

Apply `func` to every input to get its label.

Get one batch from the data loader of `ds_type`. Optionally `detach` and `denorm`.

Get `item` into a batch. Optionally `detach` and `denorm`.

Return a relu activation, maybe `leaky` and `inplace`.

Split the items randomly by putting `valid_pct` in the validation set, optional `seed` can be passed.

Reduce flatten loss to give a single loss value for each image

Find 8 coeff mentioned [here](https://web.archive.org/web/20150222120106/xenia.media.mit.edu/~cwren/interpolator/).

`sw`,`sh` scale width,height - `c`,`r` focus col,row.

Add `names` to the inner metric names.

Add the `items` as a test set. Pass along `label` otherwise label them with `EmptyLabel`.

A batchnorm2d layer with `nf` features initialized depending on `norm_type`.

Stops asynchronous request queue processing thread.

Return module from `mod_name`.

Label `self.items` from the values in `cols` in `self.inner_df`.

Read a notebook in `fname` and return its corresponding json

Read a hyperparameter `key` in the optimizer dictionary.

Resize the image to `size`, size can be a single int.

Initialize all non-batchnorm layers of `m` with `init_func`.

Get the metadata associated with `arch`.

Create a sequence of convolutional (`ni` to `nf`), ReLU (if `use_activ`) and batchnorm (if `bn`) layers.

Return predictions and targets on the valid, train, or test set, depending on `ds_type`.

Return the inner state of the `Callback`, `minimal` or not.

Check if `func` accepts `arg`.

Create a `LabelList` from `state`.

Calculate loss and metrics for a batch, call out to callbacks as necessary.

Return the representation of the first `n_max` elements in `items`.

Like `plt.subplots` but with consistent axs shape, `kwargs` passed to `fig.suptitle` with `title`

Convert `int` or `TensorImageSize` to (height,width) of an image.

Same as `nn.CrossEntropyLoss`, but flattens input and target.