---
task_categories:
- time-series-forecasting
language:
- en
tags:
- TimeSeries,
- Multimodal,
- Heterogeneous
size_categories:
- 1B<n<10B
---
# WIATS: Weather-centric Intervention-Aware Time Series Multimodal Dataset

## Data source:

- [California ISO]
  - https://www.caiso.com/todays-outlook

## Dataset Structure

The dataset is organized into the following structure:

```
|-- subdataset1

|   |-- raw_data  # Original data files
|   |-- time_series # Rule-based Imputed data files
|   |   |-- all_version_1.parquet # Time series data for each subject can be multivariate, can be in csv, parquet, etc.
|   |   |-- all_version_2.parquet
|   |   |-- ...
|   |   |-- id_info.json # Metadata for each subject

|   |-- weather
|   |   |-- location_1
|   |   |   |-- raw_data
|   |   |   |   |-- daily_weather_raw_????.json
|   |   |   |   |-- ...
|   |   |   |   |-- daily_weather_????.csv
|   |   |   |   |-- ...
|   |   |   |   |-- hourly_weather_????.csv
|   |   |   |   |-- ...
|   |   |   |-- weather_report (can be flattened and use regex to extract the version)
|   |   |   |   |-- version_1
|   |   |   |   |   |-- xxx_weather_report_????.json
|   |   |   |   |   |-- ...
|   |   |   |   |-- version_2
|   |   |   |   |-- ...
|   |   |   |-- report_embedding # embedding for the weather report
|   |   |   |   |-- version_1
|   |   |   |   |   |-- xxx_report_embedding_????.pkl
|   |   |   |   |   |-- ...
|   |   |   |   |-- version_2
|   |   |   |   |-- ...

|   |   |-- location_2
|   |   |-- ...

|   |   |-- merged_report_embedding # merged embedding for multiple needed locations (optional)
|   |   |   |-- xxx_embeddings_????.pkl
|   |   |   |-- ...

|   |   |-- merged_general_report # merged general report for multiple needed locations (optional)
|   |   |   |-- xxx_report.json
|   |   |   |-- ...

|   |-- scripts # Scripts for data processing, model training, and evaluation
|   |-- id_info.json # Metadata for whole dataset without preprocessing
|   |-- static_info.json # Static information for this dataset, including the dataset information, channel information, downtime reasons. 
|   |-- static_info_embeddings.pkl

|-- subdataset2
|-- ......

```

## id_info.json Structure

The `id_info.json` file contains metadata for each subject in the dataset. Extracted from the raw dataset. The structure is as follows:

```json
{
    "id_1": {
        "len": 1000, # Length of the time series data
        "sensor_downtime": {
            1: {
                "time": [yyyy-mm-dd hh:mm:ss, yyyy-mm-dd hh:mm:ss],
                "index": [start_index, end_index]
            },
            2: {
                "time": [yyyy-mm-dd hh:mm:ss, yyyy-mm-dd hh:mm:ss],
                "index": [start_index, end_index]
            },
            ...
        },
        "other_info_1": "value_1", # Other information about the subject customizable entry
        "other_info_2": "value_2",
        ...
    },
    "id_2": ...

}
```

## static_info.json Structure

The `static_info.json` file contains static information for the whole dataset. The structure is as follows:

```json
{
    "general_info": "description of the dataset",
    "downtime_prompt": "",
    "channel_info": {
        "id_1": {
            "channel_1": "channel 1 is xxx",
            "channel_2": "channel 2 is xxx"
        },
        "id_2": {
            "channel_1": "channel 1 is xxx",
            "channel_2": "channel 2 is xxx"
        },
        ...
    },
}
```