Create Session

Create a new session in the Console for interactive forecasting and result tracking.

Forecasting tasks are managed by “sessions”. Click “Create session” in the upper-left to start a new forecasting task.

Note: Each session stores its own data, parameters, and results. You can switch between sessions to review historical forecasts at any time.

Steps

1. Open the Console

Sign in to the Timer Console.

2. Create a session

Create a new session and pick a model and configuration.

• Click the entry for “New session / New forecast”
• Optionally name the session
• Choose model and forecast settings

3. Run a forecast

Upload data or select a data source, run the forecast, and review the outputs.

---

Add time-series data

After creating a session, add time-series data for the target variable. Choose one of the three input methods below.

Method 1: Draw a curve

Best for: Quickly validating results or when you don’t have a data file ready.

1. Open the “Draw curve” tab.
2. Click and drag on the canvas to draw a time-series curve.
3. By default, the system samples 48 points with values in the range [0, 100].
4. To redraw, click “Reset” to clear the canvas.

Try to cover the full canvas. More informative patterns generally lead to better forecasts.

Method 2: Paste or input values

Best for: When you already have numeric values and want to paste or type them directly.

1. Open the “Input data” tab.
2. Enter one record per line in the format: timestamp,value; you may also input values only (without timestamps).
3. You can also copy from spreadsheets (e.g., Excel) and paste into the textbox.

Example format:

2024-01-01 00:00:00,120
2024-01-01 01:00:00,135
2024-01-01 02:00:00,142

Each line is time,value, separated by an English comma. Use a fixed time interval whenever possible, such as hourly or daily. For timestamp format requirements, see the table and constraints below.

Format type	Supported style	Examples
ISO 8601 / RFC3339	Standard date-time with T; may include Z, a time zone offset, or fractional seconds.	2024-01-01T00:00:00Z 2024-01-01T00:00:00+08:00 2024-01-01T00:00:00.123Z
Hyphen date-time	Uses - between year, month, and day. Supports date-only, minute-level, and second-level precision; both zero-padded and non-zero-padded forms are supported.	2024-01-01 2024-01-01 00:00 2024-01-01 00:00:00 2024-1-1 8:00
Slash date-time	Uses / between year, month, and day. Supports date-only, minute-level, and second-level precision; both zero-padded and non-zero-padded forms are supported.	2024/01/01 2024/01/01 00:00 2024/01/01 00:00:00 2024/1/1 8:00
Chinese date-time	Uses Chinese year, month, and day characters. Supports date-only, minute-level, and second-level precision; both zero-padded and non-zero-padded forms are supported.	2024年01月01日 2024年01月01日 00:00 2024年01月01日 00:00:00 2024年1月1日
Unix timestamp	Uses a number or numeric string. Unix timestamps are interpreted in milliseconds by default; numeric strings are treated as timestamps, not compact date strings such as yyyyMMdd.	1704067200000 "1704067200000"

Constraints:

• Format validity: Timestamps must use one of the supported formats. Unsupported timestamp formats will cause a validation error. Convert them to a supported format before submitting again.
• Format consistency: Timestamps in the same inference batch must use a consistent format. If any two data points use inconsistent timestamp formats, the system will return a validation error. Normalize timestamp formats before submitting again.

Method 3: Upload a file

Best for: Real workflows when you have prepared CSV or TsFile files.

Supported file formats:

Format	Description	Typical use
CSV	Comma-separated text file with broad compatibility	Exports from Excel or databases
TsFile	Binary time-series format with high read/write efficiency	Exports from IoTDB and similar TSDBs

Steps:

1. Open the “Upload file” tab.
2. Click the upload area or drag the file into the dashed box to upload.
3. After upload, the system parses the file and shows basic info (row count, column names).
4. In “Assign column roles”, set each column as Time / Target / Covariate / Covariate (including future data) / Ignore.
5. Click “Confirm” to finish loading the data.

Three sample files are available at the bottom (values-only / timestamps / with covariates) to help you try the full flow quickly.

---

Add covariates (optional)

Covariates are external factors that influence the target variable, such as temperature, holidays, or promotions. Adding covariates can improve forecasting accuracy.

Depending on whether covariates include future data, the system supports three scenarios with increasing accuracy:

Scenario	Configuration	Description
Scenario 1	Target only	Forecast from historical target values only; suitable when no external factors are available.
Scenario 2	Target + covariates (no future data)	Use historical covariates to help modeling, but no covariate values are provided for the forecast horizon.
Scenario 3	Target + covariates (including future data)	Covariates are known for both history and future (e.g., holiday calendars), usually giving the best accuracy.

How to add covariates:

1. After loading the target data, expand ”+ Add covariates (optional)”.
2. Use the same three methods (draw / input / upload) to add covariate data.
3. Add multiple covariates; each one can be configured independently.
4. If covariates include future data, include them and mark the column as “Covariate (including future data)” so the model can leverage that information.

---

Forecast parameters

Click “Forecast parameters” in the upper-right to configure the following settings:

Parameter	Meaning	Range / Default	Recommendation
Steps	Number of time steps to forecast into the future	1–720, default 10	Set based on your needs
Start	Which row index to start forecasting from	Integer, defaults to the last row	Leave empty to use all historical data

If “Start” is set to a value smaller than the total rows, the system forecasts using the first “start” rows and compares the next “steps” rows with actual values for evaluation.

---

Run and view results

After data and parameters are ready, click the blue arrow button (→) to submit the forecasting task.

---

Quick try: three built-in examples

Three built-in examples are provided at the bottom to demonstrate the three input methods. Click one to load it instantly.

Example	Input method	Description
Forecast sales trend	Draw curve	Draw a historical sales curve and run a quick forecast.
Transformer oil temperature	Input data	Paste an industrial temperature series to validate the model.
Ice cream sales	Upload file	Upload a CSV with covariates (temperature) to try multivariate forecasting.

Each example comes with prefilled data. Click the example name to load it with no manual input, ideal for first-time users.

---

Model overview

Tip: If you’re unsure which model to choose, use Auto. The system will select the optimal forecasting approach for your data. For larger-scale data ingestion, consider a private deployment. Contact us for details.

---

FAQ

Q: What if column roles are auto-detected incorrectly after upload?

In “Assign column roles”, set each column to the correct role (Time / Target / Covariate / Ignore), then click “Confirm”.

Q: Can I use covariates without future data?

Yes. Mark the column as “Covariate” (instead of “Covariate (including future data)”). The system will use only historical covariate values. Accuracy may be slightly lower than with future data, but typically better than using none.

Q: Can I edit after drawing a curve?

Yes. Click “Reset” to redraw, or switch to “Input data” to edit values directly.