> For the complete documentation index, see [llms.txt](https://deeplearning4j.konduit.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://deeplearning4j.konduit.ai/en-1.0.0-rewrite/model-import/overview-1/constraints.md).

# Constraints

Keras allows weight constraints to be applied to layer parameters during training. These constraints clip or rescale the weights after each gradient update to keep them within a specified range. DL4J supports all standard Keras constraints via corresponding `LayerConstraint` implementations.

The mapping is implemented in [KerasConstraintUtils](https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-modelimport/src/main/java/org/deeplearning4j/nn/modelimport/keras/utils/KerasConstraintUtils.java).

## Supported Constraints

| Keras constraint | DL4J LayerConstraint    | Description                                  |
| ---------------- | ----------------------- | -------------------------------------------- |
| `max_norm`       | `MaxNormConstraint`     | Clips incoming weights so                    |
| `non_neg`        | `NonNegativeConstraint` | Forces all weights to be non-negative        |
| `unit_norm`      | `UnitNormConstraint`    | Scales incoming weights to have unit L2 norm |
| `min_max_norm`   | `MinMaxNormConstraint`  | Clips weights so min\_value ≤                |

## Constraint Descriptions

### max\_norm

Clips the incoming weights for each neuron so that their L2 norm does not exceed `max_value`. Applied along the `axis` dimension (default: 0, i.e., incoming connections per unit).

```python
# Keras
from keras.constraints import max_norm
layer = Dense(64, kernel_constraint=max_norm(2.0))
```

```java
// DL4J equivalent
new MaxNormConstraint(2.0, 1)
```

**Keras parameters:**

| Parameter   | Default | Description                   |
| ----------- | ------- | ----------------------------- |
| `max_value` | 2       | Maximum L2 norm value         |
| `axis`      | 0       | Axis along which to constrain |

### non\_neg

Forces all weights to be greater than or equal to zero after each update. Any negative weights are clipped to zero. Useful for non-negative matrix factorisation tasks.

```python
# Keras
from keras.constraints import non_neg
layer = Dense(64, kernel_constraint=non_neg())
```

```java
// DL4J equivalent
new NonNegativeConstraint()
```

### unit\_norm

Scales incoming weights so that their L2 norm equals 1. Effectively normalises the weight vector for each unit.

```python
# Keras
from keras.constraints import unit_norm
layer = Dense(64, kernel_constraint=unit_norm())
```

```java
// DL4J equivalent
new UnitNormConstraint(1)
```

### min\_max\_norm

Clips incoming weights so that their L2 norm is between `min_value` and `max_value`. The `rate` parameter controls the speed of clipping (rate=1.0 applies the constraint exactly; values less than 1.0 apply it more slowly).

```python
# Keras
from keras.constraints import min_max_norm
layer = Dense(64, kernel_constraint=min_max_norm(min_value=0.5, max_value=2.0))
```

```java
// DL4J equivalent
new MinMaxNormConstraint(0.5, 2.0, 1.0, 1)
```

**Keras parameters:**

| Parameter   | Default | Description                             |
| ----------- | ------- | --------------------------------------- |
| `min_value` | 0.0     | Minimum L2 norm value                   |
| `max_value` | 1.0     | Maximum L2 norm value                   |
| `rate`      | 1.0     | Rate at which the constraint is applied |
| `axis`      | 0       | Axis along which to constrain           |

## Usage Example

```java
import org.deeplearning4j.nn.modelimport.keras.KerasModelImport;
import org.deeplearning4j.nn.multilayer.MultiLayerNetwork;

// Constraints are automatically read and applied to the imported network configuration
MultiLayerNetwork model = KerasModelImport.importKerasSequentialModelAndWeights("model.h5");
```

## Notes

* Constraints are only relevant when the imported model is used for **continued training** in DL4J. For inference, constraints have no effect.
* The constraint `axis` parameter from Keras is mapped to the equivalent DL4J axis convention. Verify constraint behaviour when porting models with non-default axis values.
* Custom Python constraint classes cannot be imported.


---

# 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 dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://deeplearning4j.konduit.ai/en-1.0.0-rewrite/model-import/overview-1/constraints.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
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.