TimeGEN-1 Quickstart (Azure)

Quickstart guide to deploy and use TimeGEN-1 on Azure with the Nixtla Python SDK for time series forecasting.

TimeGEN-1 is TimeGPT optimized for Azure infrastructure. It is a production-ready generative pretrained transformer for time series, capable of accurately predicting domains such as retail, electricity, finance, and IoT with minimal code.

Key Benefit

Azure-native generative forecasting with TimeGEN-1 for streamlined deployments.

Use Cases

• Demand forecasting\

• Electricity load prediction\

• Financial time series\

• IoT data analysis

1. Step 1: Set up a TimeGEN-1 endpoint on Azure and generate an API key

   1. Visit ml.azure.com and sign in (or create a Microsoft account if needed).
   2. Click Models in the sidebar.
   3. Search for TimeGEN in the catalog and select TimeGEN-1.
   4. Click Deploy to create an endpoint.

   

   TimeGEN-1 model catalog deployment option.

   5. Click Endpoint in the sidebar.
   6. Copy the base URL and API Key shown for your TimeGEN-1 endpoint.

   

   Endpoint URL and API key for TimeGEN-1.
2. Step 2: Install Nixtla Python SDK

   Install the nixtla package using pip:

   Install nixtla SDK

   pip install nixtla
   

3. Step 3: Import and instantiate the Nixtla client

   Import the Nixtla client into your Python environment:

   Import NixtlaClient

   from nixtla import NixtlaClient
   

   Then create a client instance using your TimeGEN-1 endpoint credentials:

   Instantiate NixtlaClient

   nixtla_client = NixtlaClient(
       base_url="YOUR_BASE_URL",
       api_key="YOUR_API_KEY"
   )
   

4. Step 4: Load your time series data

   In this example, we’ll use the classic AirPassengers dataset to demonstrate forecasting. The dataset shows monthly passenger counts in Australia between 1949 and 1960.

   Load AirPassengers dataset

   import pandas as pd
   
   df = pd.read_csv(
       'https://raw.githubusercontent.com/Nixtla/transfer-learning-time-series/main/datasets/air_passengers.csv'
   )
   df.head()
   

   Use the Nixtla client to quickly visualize your data:

   Visualize time series

   nixtla_client.plot(df, time_col='timestamp', target_col='value')
   

   

   AirPassengers time series sample visualized.

   Data Requirements

   • Ensure the target column has no missing or non-numeric values.\

   • Avoid gaps in date stamps (for the specific frequency) from the initial to final timestamp—missing dates are not automatically imputed.\

   • Datestamps must be in a pandas-readable format. (See Pandas reference)

   See Data Requirements for details.

   Saving Figures

   In most notebook environments, figures display automatically. To save a figure locally, run:

   Save plot figure

   fig = nixtla_client.plot(df, time_col='timestamp', target_col='value')
   fig.savefig('plot.png', bbox_inches='tight')
   

5. Step 5: Generate forecasts

   Use the forecast method from the Nixtla client to forecast the next 12 months.

   Parameters

   • df: Pandas DataFrame with time series data\

   • h: Forecast horizon (number of steps ahead)\

   • freq: Time series frequency (pandas frequency aliases)\

   • time_col: Name of timestamp column\

   • target_col: Name of forecast variable

   Generate 12-month forecast

   timegen_fcst_df = nixtla_client.forecast(
       df=df,
       h=12,
       freq='MS',
       time_col='timestamp',
       target_col='value'
   )
   timegen_fcst_df.head()
   

   Forecast endpoint call logs will be displayed for validation and preprocessing steps.

   Forecast API Call Log

   Forecast API call logs

   INFO:nixtla.nixtla_client:Validating inputs...
   INFO:nixtla.nixtla_client:Preprocessing dataframes...
   INFO:nixtla.nixtla_client:Inferred freq: MS
   INFO:nixtla.nixtla_client:Restricting input...
   INFO:nixtla.nixtla_client:Calling Forecast Endpoint...
   

   Example output:

   	timestamp	TimeGPT
   0	1961-01-01	437.837921
   1	1961-02-01	426.062714
   2	1961-03-01	463.116547
   3	1961-04-01	478.244507
   4	1961-05-01	505.646484

   Visualize the forecast results:

   Visualize forecast results

   nixtla_client.plot(df, timegen_fcst_df, time_col='timestamp', target_col='value')
   

   

   Forecast visualization for the AirPassengers dataset.