> For the complete documentation index, see [llms.txt](https://docs.zingg.ai/zingg/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.zingg.ai/zingg/running-zingg/create-training-data/label-training-pairs.md).

# Label Training Pairs

Review candidate pairs and label each as Match, No Match, or Can't Say. The human feedback step that teaches Zingg what a match looks like in your data.

The `label` phase opens an interactive layer where you review the candidate pairs found by `findTrainingData` and mark each pair. This is the only step in the Zingg workflow that requires human input. No ML knowledge is needed, just your domain understanding of whether two records represent the same real-world entity.

If you already have labeled data from an external source, you can supply it directly using `trainingSamples` in your configuration. See [Create Training Data](/zingg/running-zingg/create-training-data.md) for how to set this up.

30 to 40 matching pairs is a strong starting point. Label until you feel that your labeled examples represent all field types and data patterns in your schema. If accuracy needs improvement after your first match run, return to labelling, focus on the patterns or field combinations that appear to be missing or underrepresented.

{% stepper %}
{% step %}

### Step 1: Label

Review each pair presented. Enter your decision: Match (1), No Match (2), or Can't Say (0).
{% endstep %}

{% step %}

### Step 2: Save

Your labels are saved automatically to `zinggDir/modelId` after each session.
{% endstep %}

{% step %}

### Step 3: Iterate

if the number of matched and unmatched pairs are still not sufficient, then iterate. Run `findTrainingData` again to obtain a fresh set of candidate pairs. Label those. Repeat until Zingg's predictions align with your expectations.
{% endstep %}
{% endstepper %}

{% tabs %}
{% tab title="Community" %}

### **Python**

```python
options = ClientOptions([ClientOptions.PHASE, "label"])
zingg = Zingg(args, options)
zingg.initAndExecute()
```

### **CLI**

```bash
./scripts/zingg.sh --phase label --conf config.json --showConcise=true
```

{% hint style="success" icon="right-long" %}
`--showConcise=true` is optional. It only shows fields which are not `DONT_USE`, making the labelling session cleaner when you have many fields.
{% endhint %}

{% hint style="info" icon="right-long" %}
Need to correct a pair you already labeled? See [Update Label](/zingg/running-zingg/update-label.md).
{% endhint %}
{% endtab %}

{% tab title="Enterprise" %}

#### Python

```python
options = ClientOptions([ClientOptions.PHASE, "label"])
zingg = EZingg(args, options)
zingg.initAndExecute()
```

### **CLI**

```bash
./scripts/zingg.sh --phase label --conf config.json --showConcise=true
```

{% hint style="success" icon="right-long" %}
`--showConcise=true` is optional. It only shows fields which are not `DONT_USE`, making the labelling session cleaner when you have many fields.
{% endhint %}

{% hint style="info" icon="right-long" %}
Notebook provides a visual widget showing one pair at a time with Match, No Match, and Can't Say buttons.
{% endhint %}
{% endtab %}

{% tab title="Enterprise Snowflake" %}

### Run label

Zingg on Snowflake can be run either from a local terminal via the CLI, or natively inside Snowflake using an interactive labeling service.

#### CLI (local terminal)

{% hint style="success" icon="right-long" %}
`--showConcise=true` is optional. It only shows fields which are not `DONT_USE`, making the labelling session cleaner when you have many fields.
{% endhint %}

```bash
./scripts/zingg.sh --phase label --conf config.json --showConcise=true \
--properties-file <location to snowflake.properties>
```

#### Snowflake (interactive service)

The `label` phase is interactive, so instead of an async job, it runs as a standing service you connect to directly.

```sql
CREATE SERVICE CONTAINER_ZINGG_DB.PUBLIC.ZINGG_CLI_SERVICE
IN COMPUTE POOL CONTAINER_ZINGG_POOL
FROM @specs
SPECIFICATION_FILE = '<zingg-cli-spec-file>'
EXTERNAL_ACCESS_INTEGRATIONS = (ALLOW_ALL_EAI);
```

`<zingg-cli-spec-file>` is the specification file that defines the container and service configuration used to run the Zingg CLI interactively inside Snowpark Container Services. It is maintained by your administrator and referenced from your `@specs` stage.

Once the service is running, retrieve its endpoint to connect to the labeling session.

```sql
SHOW ENDPOINTS IN SERVICE ZINGG_CLI_SERVICE;
```

Use the returned endpoint to open the interactive labeling terminal and label pairs as you would via CLI.

**Stop the compute pool**

Once you're done labeling, stop the compute pool to release resources.

```sql
ALTER COMPUTE POOL CONTAINER_ZINGG_POOL STOP ALL;
```

{% hint style="info" icon="right-long" %}
Notebook provides a visual widget showing one pair at a time with Match, No Match, and Can't Say buttons.
{% endhint %}
{% endtab %}
{% endtabs %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.zingg.ai/zingg/running-zingg/create-training-data/label-training-pairs.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.