# fail_on_no_train_batches

Python API: `luxonis_train.callbacks.fail_on_no_train_batches`

Stops a fit with an error when the training loader yields no batches.

The error message shows the minimum dataset size for the current settings.

## Classes

### FailOnNoTrainBatches

Callback that stops a fit that has no training batches.

A fit has no training batches when the training dataset is empty. It also has none when the dataset is smaller than the effective
batch size and the loader drops the last incomplete batch.
[LuxonisModel](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/core/core.md)
drops it when `trainer.skip_last_batch` is set. Without this callback, Lightning stops the fit and only logs an info message.

[LuxonisModel](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/core/core.md)
gives this callback to its trainer. The config does not have to list it.

#### Methods

##### on_fit_start

```python
def on_fit_start(trainer: pl.Trainer, pl_module: lxt.LuxonisLightningModule):
```

Raise an error when the fit loop has no training batches.

Lightning calls this hook at the start of `trainer.fit`. The hook sets up the training data of the fit loop, so that Lightning
computes the number of training batches early. When that number is `0`, the hook raises an error.

The message of the error shows these values when the hook can find them:

 * the size of the dataset;
 * the minimum dataset size that the current settings need;
 * the number of missing samples, when the dataset is smaller than that minimum;
 * the batch size, the world size, `drop_last` of the loader, and `limit_train_batches` of the trainer.

The hook reads the batch size from the loader, or from `trainer.batch_size` of the config.

Parameters

 * `trainer` (`pl.Trainer`): The trainer of the fit.
 * `pl_module` (`lxt.LuxonisLightningModule`): The model. The hook reads its config.

Raises

 * `RuntimeError`: When the fit loop has no training batches.