> ## Documentation Index
> Fetch the complete documentation index at: https://docs.canaryedge.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Fine-Tuning

> Improve detection accuracy with per-machine baselines

Canary Edge ships with a powerful generic model that works out of the box. For production workloads, setting a **baseline** for each machine fine-tunes a dedicated predictor that learns the specific patterns of your equipment.

## When to Set a Baseline

* Your time series has unique seasonal patterns
* The generic model produces too many false positives
* You need maximum sensitivity for critical equipment
* You want regime classification (HEALTHY/ACTIVE/TRANSITION/SHOCK)

## How It Works

1. Send normal operating data via `POST /v1/baseline` with a `machine_id`
2. Canary Edge computes energy statistics and trains a lightweight predictor (\~462K params) in seconds
3. Future detection calls with that `machine_id` use the fine-tuned model automatically
4. Detection accuracy typically improves from \~82% (generic) to 97%+ (fine-tuned)

## Setting a Baseline

<CodeGroup>
  ```bash cURL theme={null}
  curl -X POST https://api.canaryedge.com/v1/baseline \
    -H "Content-Type: application/json" \
    -H "Ocp-Apim-Subscription-Key: YOUR_API_KEY" \
    -d '{
      "machine_id": "pump-assembly-1",
      "energies": [0.12, 0.15, 0.11, 0.13, 0.14, 0.12, 0.16, 0.11, ...]
    }'
  ```

  ```python Python theme={null}
  import requests

  # Step 1: Collect energy scores from normal operation
  energies = []
  for window in normal_operation_windows:
      resp = requests.post(
          "https://api.canaryedge.com/v1/detect",
          headers={"Ocp-Apim-Subscription-Key": "YOUR_API_KEY"},
          json={"machine_id": "pump-assembly-1", "series": window}
      )
      energies.append(resp.json()["energy"])

  # Step 2: Set the baseline
  requests.post(
      "https://api.canaryedge.com/v1/baseline",
      headers={"Ocp-Apim-Subscription-Key": "YOUR_API_KEY"},
      json={"machine_id": "pump-assembly-1", "energies": energies}
  )
  ```
</CodeGroup>

<Warning>
  Only send energy scores from **normal operating data** for baseline creation. Including anomalous data will degrade detection accuracy.
</Warning>

## Checking a Baseline

```bash theme={null}
curl "https://api.canaryedge.com/v1/baseline?machine_id=pump-assembly-1" \
  -H "Ocp-Apim-Subscription-Key: YOUR_API_KEY"
```

Response:

```json theme={null}
{
  "machine_id": "pump-assembly-1",
  "mean": 0.132,
  "std": 0.018,
  "p99": 0.187,
  "sample_count": 500,
  "created_at": "2026-03-15T10:00:00Z"
}
```

## Requirements

* Minimum **100 energy values** for a reliable baseline
* Data should represent at least one full operational cycle
* All data should be from normal, healthy operation

## Machine Status

Once a baseline is set, check machine status:

```bash theme={null}
curl "https://api.canaryedge.com/v1/status/pump-assembly-1" \
  -H "Ocp-Apim-Subscription-Key: YOUR_API_KEY"
```

## Deleting a Machine

To remove a baseline and revert to the generic detector:

```bash theme={null}
curl -X DELETE "https://api.canaryedge.com/v1/machines/pump-assembly-1" \
  -H "Ocp-Apim-Subscription-Key: YOUR_API_KEY"
```