# Heavy Decoder Autoencoder

The heavy decoder autoencoder has a higher number of layers in the decoder as compared to the encoder. This design choice can be useful where capturing intricate details during data reconstruction is important. 

```python
from sdgne.datagenerator.autoencoder import HeavyDecoderAutoencoder

minority_column_label = 'class'
minority_class_label = 0

synthesizer = HeavyDecoderAutoencoder(dataset,
                               minority_column_label,
                               minority_class_label)
```

## Importing Heavy Decoder Autoencoder

```python
from sdgne.datagenerator.autoencoder import HeavyDecoderAutoencoder
```

***

## Creating a synthesizer

```python
synthesizer = HeavyDecoderAutoencoder(dataset,
                               minority_column_label,
                               minority_class_label)
```

### Parameters

<table data-header-hidden data-full-width="false"><thead><tr><th width="216.66666666666669">variable</th><th width="140">type</th><th>datatype</th><th>info</th></tr></thead><tbody><tr><td>dataset</td><td>required</td><td>pd.Dataframe</td><td>Represents a pandas data frame containing both, the original minority and the original majority data</td></tr><tr><td>minority_column_label</td><td>required</td><td>string</td><td>Represents the column label. Eg. 'class' , 'output'</td></tr><tr><td>minority_class_label</td><td>required</td><td>string</td><td>Represents the minority class label. Eg. '1' , '0'</td></tr></tbody></table>

### Returns

An instance of class HeavyDecoderAutoencoder.

## Network Architecture

Below is the network architecture for the Heavy Decoder Autoencoder. Here, the encoder has two layers, with 22 and 20 nodes respectively. The bottleneck is a single Dense layer with 16 nodes. The decoder has four layers with 18,20, 22, and 24 nodes respectively. The decoder layer activation is sigmoid.

| encoder\_dense\_layers | 22, 20         |
| ---------------------- | -------------- |
| bottle\_neck           | 16             |
| decoder\_dense\_layers | 18, 20, 22, 24 |
| decoder\_activation    | sigmoid        |

## Synthetic Data Generation

### data\_generator()

The data generator function generates synthetic data using the synthesizer. It has an option parameter  `no_of_syntetic_data.`

* If no\_of\_syntetic\_data is not defined, data\_generator by default generates the \`n\` number of synthetic data such that the majority data and minority datasets get balanced.
* If no\_of\_syntetic\_data is defined and the dataset is already balanced, data\_generator generates \
  2 \* Number of original minority data.
* If no\_of\_syntetic\_data is defined and the dataset is not balanced, it generates synthetic data equal to the value passed.

```python
synthesized_data = synthesizer.data_generator(no_of_syntetic_data)
```

### Parameters

<table data-header-hidden><thead><tr><th width="204">variable</th><th width="212">type</th><th>datatype</th><th>info</th></tr></thead><tbody><tr><td>no_of_syntetic_data</td><td>optional (default: None)</td><td>integer</td><td>Represents the number of synthetic data to be generated</td></tr></tbody></table>

### Returns

A pandas data frame that combines original data and synthetic data.

## Usage

`synthesizer.data_generator()`

`synthesizer.data_generator(no_of_syntetic_data=100)`

{% hint style="warning" %}
synthesize\_data returned from data\_generator( ) adds a column \`**synthetic\_data**\` to the data frame.&#x20;

\
df\['synthetic\_data'] = 0 : For original data\
df\['synthetic\_data'] = 1  : For Synthetic generated data&#x20;
{% endhint %}