zaai-ai/hierarchical_time_series_datasets

---
language:
- en
license:
- cc-by-4.0
size_categories:
- 10K<n<100k
task_categories:
- time-series-forecasting
task_ids:
- univariate-time-series-forecasting
- multivariate-time-series-forecasting
---

# Dataset Repository

This repository includes several datasets: **Houston Crime Dataset**, **Tourism in Australia**, **Prison in Australia**, and **M5**. These datasets consist of time series data representing various metrics across different categories and groups.

## Dataset Structure

Each dataset is divided into training and prediction sets, with features such as groups, indices, and time series data. Below is a general overview of the dataset structure:

### Training Data
The training data contains time series data with the following structure:
- **x_values**: List of time steps.
- **groups_idx**: Indices representing different group categories (e.g., Crime, Beat, Street, ZIP for Houston Crime).
- **groups_n**: Number of unique values in each group category.
- **groups_names**: Names corresponding to group indices.
- **n**: Number of time series.
- **s**: Length of each time series.
- **n_series_idx**: Indices of the time series.
- **n_series**: Indices for each series.
- **g_number**: Number of group categories.
- **data**: Matrix of time series data.

### Prediction Data
The prediction data has a similar structure to the training data and is used for forecasting purposes.

**Note:** It contains the complete data, including training and prediction sets.

### Additional Metadata
- **seasonality**: Seasonality of the data.
- **h**: Forecast horizon.
- **dates**: Timestamps corresponding to the time steps.

## Example Usage

Below is an example of how to load and use the datasets using the `datasets` library:

```python
from datasets import load_dataset

# Load the datasets from Hugging Face Hub
m5_data = load_dataset('zaai-ai/hierarchical_time_series_datasets', 'm5')
police_data = load_dataset('zaai-ai/hierarchical_time_series_datasets', 'police')
prison_data = load_dataset('zaai-ai/hierarchical_time_series_datasets', 'prison')
tourism_data = load_dataset('zaai-ai/hierarchical_time_series_datasets', 'tourism')

# Example: Accessing specific data from the datasets
print("M5 Data:", m5_data)
print("Police Data:", police_data)
print("Prison Data:", prison_data)
print("Tourism Data:", tourism_data)

# Access the training data
train_data = prison_data['train']

# Access the prediction data
predict_data = prison_data['predict']

# Example: Extracting x_values and data
x_values = train_data['x_values']
data = train_data['data']

print(f"x_values: {x_values}")
print(f"data shape: {data.shape}")
```

### Steps to Follow:

1. **Install the datasets library:**
```sh
pip install datasets
```
2. **Load the Datasets:**
- Use the load_dataset function from the datasets library to load the datasets directly from the Hugging Face Hub.