Method __getitem__

loaders/custom_dataset.py:266–319

(self, index)

        self.cache_indices[self.current_cache_index] = dataset_index

    def __getitem__(self, index):
        try:
            dataset_index = 0
            while index >= self.cumulative_lengths[dataset_index + 1]:
                dataset_index += 1

            if self.cache_indices[self.current_cache_index] != dataset_index:
                # Cache the dataset if it's not already cached.
                self.cache_dataset(dataset_index)

            # Retrieve the un-cumulative index of the current sample.
            start_index = (
                index
                if dataset_index == 0
                else index - self.cumulative_lengths[dataset_index]
            )

            if self.lighten:
                # If the "lighten" option is enabled, we use only the first 5 levels of the orderbook (i.e. 4_level_features * 5_levels = 20_orderbook_features).
                window_data = self.cache_data[self.current_cache_index][
                    start_index: start_index + self.window_size, :20
                ]
            else:
                # If the "lighten" option is not enabled, we use all the 10 levels of the orderbook (i.e. 4_level_features * 10_levels = 40_orderbook_features).
                window_data = self.cache_data[self.current_cache_index][
                    start_index: start_index + self.window_size, :40
                ]

            # Determine the position of the prediction horizon in the list of all horizons.
            position = next(
                (
                    index
                    for index, value in enumerate(self.all_horizons)
                    if value == self.prediction_horizon
                ),
                None,
            )
            # Extract the label from the dataset given its position.
            label = self.cache_data[self.current_cache_index][
                start_index + self.window_size, 40:
            ][position]
            # Discretize the label using the provided threshold.
            if self.backtest is False:
                if label > self.threshold:
                    label = 2
                elif label < -self.threshold:
                    label = 0
                else:
                    label = 1

            return torch.tensor(window_data).unsqueeze(0), torch.tensor(label)
        except Exception as e:
            print(f"Exception in DataLoader worker: {e}")
            raise e


''&#x27;