Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
multiple_series.mdx 3.76 KB
Newer Older
bailuo's avatar
readme  
bailuo committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85 
---
title: "Multiple Time Series"
description: "Learn how to handle missing values in time series data for accurate forecasting with TimeGPT."
icon: "table"
---

You can pass multiple time series within the same dataset to TimeGPT. We can then make forecasts or detect anomalies on all series simultaneously.

To include multiple series, simply include a unique identifier column. By default, we expect this column to be called `unique_id`. The identifier column assigns a value to each series such that we can distinguish between them.

## Load Data with Multiple Series

Here is an example of loading a dataset with multiple series inside. 

```python
df = pd.read_csv('https://raw.githubusercontent.com/Nixtla/transfer-learning-time-series/main/datasets/electricity-short-with-ex-vars.csv')
df['ds'] = pd.to_datetime(df['ds'])
df = df[["unique_id", "ds", "y"]]
df.groupby('unique_id').head(1)
```

<Frame caption="Multiple-Series Data Preview">

| unique_id | ds	     | y      |
| --------- | ---------- | ------ |
| BE	    | 2016-10-22 |	70.00 |
| DE	    | 2017-10-22 |	19.10 |
| FR	    | 2016-10-22 |	54.70 |
| NP	    | 2018-10-15 |	2.17  |

</Frame>

Above, we can see that we have four unique series in the dataset, as there are four different values in `unique_id`. Note that each series can start at different dates.

To forecast mutliple series, we can simply call:

```python Multiple Series Forecast Example
fcst = nixtla_client.forecast(df=df, h=24)
fcst.head()
```

TimeGPT will produce forecasts for all unique IDs in your DataFrame simultaneously.

### Specifying the series identifier column

In the case where unique identifier is not stored in a column called `unique_id`, you can specify the name of the column when making a call to TimeGPT:

```python Specify the name of the column for the series identifier
fcst = nixtla_client.forecast(df=df, h=24, id_col="your_column_name")
fcst.head()
```
---

## Exogenous Variables

TimeGPT supports the use of exogenous features. These are variables that are not part of the series you are trying to forecast. 

For example, suppose that you are forecasting electricity consumption, which is affected by the temperature outside. In this case, the temperature is an exogenous feature, meaning that you want to use the information from the temperature to forecast the electricity consumption.

In such case, exogenous features can be included as new columns in the dataset. Any additional column to the standard `unique_id`, `ds`, `y` format is considered as an exogenous feature.

Here is an example of loading a dataset with multiple series inside and exogenous features. 

```python Multiple Series Data Loading
df = pd.read_csv('https://raw.githubusercontent.com/Nixtla/transfer-learning-time-series/main/datasets/electricity-short-with-ex-vars.csv')
df['ds'] = pd.to_datetime(df['ds'])
df.groupby('unique_id').head(1)
```

<Frame caption="Multiple-Series with Exogenous Features Preview">

| unique_id | ds | y | Exogenous1 | Exogenous2 | day_0 | day_1 | day_2 | day_3 | day_4 | day_5 | day_6 |
|-----------|----|----|------------|------------|-------|-------|-------|-------|-------|-------|-------|
| BE | 2016-10-22 | 70.00 | 57253.00 | 49593 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 1.0 | 0.0 |
| DE | 2017-10-22 | 19.10 | 16972.75 | 15779 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 1.0 |
| FR | 2016-10-22 | 54.70 | 57253.00 | 49593 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 1.0 | 0.0 |
| NP | 2018-10-15 | 2.17 | 34078.00 | 1791 | 1.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |

</Frame>

Above, we can see that we have the columns from `Exogenous1` to `day_6` will be considered as exogenous features when forecasting with TimeGPT. 

For more information on forecasting with exogenous features, read the [Exogenous Variables tutorial](/forecasting/exogenous-variables/numeric_features) for further details.

---