### Run FuxiCTR Demo Examples

Source: https://github.com/reczoo/fuxictr/blob/main/README.md

Execute the provided demo scripts to understand basic FuxiCTR usage and workflow. Navigate to the demo directory and run the Python scripts.

```bash
cd demo
python example1_build_dataset_to_parquet.py
python example2_DeepFM_with_parquet_input.py
```

--------------------------------

### Install Requirements

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/FinalNet/README.md

Install the necessary Python packages for FinalNet by running this command.

```bash
pip install -r requirements.txt
```

--------------------------------

### Run a Model on Tiny Data (DCN Example)

Source: https://github.com/reczoo/fuxictr/blob/main/README.md

Demonstrates how to run a specific model, DCN, on a small dataset. Users can adapt this by changing the model name and modifying configuration files for custom datasets or hyperparameters. Refer to the model's specific README for more details.

```bash
cd model_zoo/DCN/DCN_torch
python run_expid.py --expid DCN_test --gpu 0
```

```bash
# Change `MODEL` according to the target model name
cd model_zoo/MODEL
python run_expid.py --expid MODEL_test --gpu 0
```

--------------------------------

### Tune Hyper-parameters with Multiple GPUs

Source: https://github.com/reczoo/fuxictr/blob/main/README.md

Utilize FuxiCTR's fast grid search functionality for hyper-parameter tuning across multiple GPUs. This example shows setting up 8 experiments using 4 GPUs.

```bash
cd experiment
python run_param_tuner.py --config config/DCN_tiny_parquet_tuner_config.yaml --gpu 0 1 2 3 0 1 2 3
```

--------------------------------

### Optional Regularization Examples

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/FinalNet/README.md

Illustrates how to specify different types of regularization for embedding matrices and network parameters. Supports L2, L1, and combined L1_L2 regularization with custom weights.

```text
embedding_regularizer: "l2(1.e-3)"
# or
embedding_regularizer: "l1(1.e-3)"
# or
embedding_regularizer: "l1_l2(1.e-3, 1.e-3)"
```

```text
net_regularizer: "l2(1.e-3)"
# or
net_regularizer: "l1(1.e-3)"
# or
net_regularizer: "l1_l2(1.e-3, 1.e-3)"
```

--------------------------------

### Model Configuration Parameters

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/FinalMLP/README.md

Configuration settings for initializing and training FuxiCTR models.

```APIDOC
## Configuration Parameters

### Description
Defines the hyperparameters and training settings for the FuxiCTR model.

### Parameters
#### Request Body
- **fs_hidden_units** (list) - Optional - Hidden units of fs gates (default: [64])
- **fs1_context** (list) - Optional - Conditional features for feature gating in stream 1 (default: [])
- **fs2_context** (list) - Optional - Conditional features for feature gating in stream 2 (default: [])
- **num_heads** (int) - Optional - Number of heads used for bilinear fusion (default: 1)
- **epochs** (int) - Optional - Max number of epochs for training (default: 100)
- **shuffle** (bool) - Optional - Whether to shuffle data samples (default: True)
- **seed** (int) - Optional - Random seed for reproducibility (default: 2021)
- **monitor** (str/dict) - Optional - Metrics for early stopping (default: 'AUC')
- **monitor_mode** (str) - Optional - 'max' or 'min' for monitor metric (default: 'max')
- **model_root** (str) - Optional - Directory to save checkpoints and logs (default: './checkpoints/')
- **early_stop_patience** (int) - Optional - Patience for early stopping (default: 2)
- **save_best_only** (bool) - Optional - Whether to save only the best model (default: True)
- **eval_steps** (int/None) - Optional - Evaluation interval (default: None)
```

--------------------------------

### Run Experiment with ShareBottom Model

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/multitask/ShareBottom/README.md

Execute the ShareBottom model test using the run_expid.py script. Specify the experiment ID and GPU to use.

```bash
python run_expid.py --expid ShareBottom_test --gpu 0 
```

--------------------------------

### Run Experiment Command

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/multitask/PLE/README.md

Use this command to execute an experiment with a specified experiment ID and GPU.

```bash
python run_expid.py --expid PLE_test --gpu 0 
```

--------------------------------

### FuxiCTR Model Configuration Parameters

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/multitask/PLE/README.md

Detailed list of parameters for configuring FuxiCTR models.

```APIDOC
## FuxiCTR Model Configuration Parameters

### Description
This section details the various parameters available for configuring FuxiCTR models, including network architecture, training settings, and model saving options.

### Parameters

#### Network Architecture
- **expert_hidden_units** (list) - Default: `[512, 256, 128]` - Hidden units in expert networks.
- **gate_hidden_units** (list) - Default: `[128, 64]` - Hidden units in gating networks.
- **tower_hidden_units** (list) - Default: `[128, 64]` - Hidden units in tower networks.
- **hidden_activations** (str/list) - Default: `"relu"` - Activation function in DNN. Layer-wise activations can be specified as a list, e.g., `["relu", "leakyrelu", "sigmoid"]`.
- **net_dropout** (float) - Default: `0` - Dropout rate in DNN.
- **batch_norm** (bool) - Default: `False` - Whether to use Batch Normalization in DNN.

#### Training Settings
- **epochs** (int) - Default: `50` - The maximum number of epochs for training, with early stopping based on monitor metrics.
- **shuffle** (bool) - Default: `True` - Whether to shuffle the data samples for each epoch of training.
- **seed** (int) - Default: `2023` - The random seed used for reproducibility.
- **monitor** (str/dict) - Default: `'AUC'` - The monitor metric for early stopping. Supports a single metric (e.g., `"AUC"`) or multiple metrics using a dict (e.g., `{"AUC": 2, "logloss": -1}` which represents `2*AUC - logloss`).
- **monitor_mode** (str) - Default: `'max'` - `'max'` means higher is better, while `'min'` denotes lower is better.
- **num_workers** (int) - Default: `3` - The number of workers for the data loader.
- **verbose** (int) - Default: `1` - `0` for silent logging, `1` for verbose logging with `tqdm`.
- **early_stop_patience** (int) - Default: `2` - Training is stopped when the monitor metric fails to improve for `early_stop_patience` consecutive evaluation intervals.

#### Model Saving
- **model_root** (str) - Default: `'./checkpoints/'` - The directory to save model checkpoints and running logs.
- **pickle_feature_encoder** (bool) - Default: `True` - Whether to pickle the feature encoder during preprocessing. Used when `data_format="csv"`.
- **save_best_only** (bool) - Default: `True` - Whether to save only the best model checkpoint.

### Request Example
```json
{
  "expert_hidden_units": [512, 256, 128],
  "gate_hidden_units": [128, 64],
  "tower_hidden_units": [128, 64],
  "hidden_activations": "relu",
  "net_dropout": 0.0,
  "batch_norm": false,
  "epochs": 50,
  "shuffle": true,
  "seed": 2023,
  "monitor": "AUC",
  "monitor_mode": "max",
  "model_root": "./checkpoints/",
  "num_workers": 3,
  "verbose": 1,
  "early_stop_patience": 2,
  "pickle_feature_encoder": true,
  "save_best_only": true
}
```

### Response

#### Success Response (200)
- **message** (str) - A confirmation message indicating successful configuration or training start.

#### Response Example
```json
{
  "message": "Model configuration applied successfully."
}
```
```

--------------------------------

### Execute Model Experiment

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/GDCN/README.md

Command to run the model using the specified experiment ID and GPU device.

```bash
python run_expid.py --expid GDCNP_test --gpu 0
```

--------------------------------

### Run FuxiCTR Experiment

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/multitask/MMoE/README.md

Command to execute an experiment using the FuxiCTR run script. Specify the experiment ID and the GPU to use.

```bash
python run_expid.py --expid MMoE_test --gpu 0 
```

--------------------------------

### Project Directory Structure

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/GDCN/README.md

Visual representation of the project file hierarchy.

```text
├── config                        # 配置文件夹
│   ├── dataset_config.yaml       # 数据集配置文件
│   └── model_config.yaml         # 模型配置文件
├── src                           # 模型代码文件夹
│   └── GDCN.py                    # 模型代码
├── fuxictr_version.py            # fuxictr加载及版本检查文件
├── README.md                     # 使用说明
├── requirements.txt              # 依赖文件
└── run_expid.py                  # 执行脚本文件
```

--------------------------------

### QNN Model Configuration Parameters

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/QNN/README.md

Defines the configuration parameters for the QNN model. These include settings for the model name, dataset, loss function, metrics, task type, optimizer, learning rate, regularization, batch size, embedding dimensions, and network architecture.

```python
model: str = "QNN_alpha"
dataset_id: str = "TBD"
loss: str = "binary_crossentropy"
metrics: list = ['logloss', 'AUC']
task: str = "binary_classification"
optimizer: str = "adam"
learning_rate: float = 1.0e-3
embedding_regularizer: float | str = 0
net_regularizer: float | str = 0
batch_size: int = 10000
embedding_dim: int = 32
num_layers: int = 3
num_row: int = 3
net_dropout: float = 0
batch_norm: bool = False
num_heads: int = 1
```

--------------------------------

### FuxiCTR Project Directory Structure

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/multitask/MMoE/README.md

Overview of the FuxiCTR project's file and directory organization. This structure helps in understanding where different components like configurations, model code, and execution scripts are located.

```text
├── config                        # 配置文件夹
│   ├── dataset_config.yaml       # 数据集配置文件
│   └── model_config.yaml         # 模型配置文件
├── src                           # 模型代码文件夹
│   └── MMoE.py                    # 模型代码
├── fuxictr_version.py            # fuxictr加载及版本检查文件
├── README.md                     # 使用说明
└── run_expid.py                  # 执行脚本文件
```

--------------------------------

### FinalNet Configuration Parameters

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/FinalNet/README.md

Defines the configuration options for the FinalNet model. This includes settings for dataset, loss, metrics, task type, optimizer, regularization, batch size, embedding dimensions, and network block specifics.

```text
model: "FinalNet"
dataset_id: "TBD"
loss: "binary_crossentropy"
metrics: ['logloss', 'AUC']
task: "binary_classification"
optimizer: "adam"
learning_rate: 1.0e-3
embedding_regularizer: 0
net_regularizer: 0
batch_size: 10000
embedding_dim: 32
block1_hidden_units: [64, 64, 64]
block1_hidden_activations: "relu"
block2_hidden_units: [64, 64, 64]
block2_hidden_activations: "relu"
block1_dropout: 0
block2_dropout: 0
```

--------------------------------

### Model Configuration (Base and Specific)

Source: https://github.com/reczoo/fuxictr/blob/main/docs/configurations.md

Defines model parameters, including shared base settings and experiment-specific configurations like DeepFM. Base settings can be overridden by expid settings.

```yaml
# model_config.yaml
Base: 
    model_root: '../checkpoints/'
    workers: 3
    verbose: 1
    patience: 2
    pickle_feature_encoder: True
    use_hdf5: True
    save_best_only: True
    every_x_epochs: 1
    debug: False
    partition_block_size: -1

DeepFM_test:
    model: DeepFM
    dataset_id: taobao_tiny_data # each expid corresponds to a dataset_id
    loss: 'binary_crossentropy'
    metrics: ['logloss', 'AUC']
    task: binary_classification
    optimizer: adam
    hidden_units: [64, 32]
    hidden_activations: relu
    net_regularizer: 0
    embedding_regularizer: 1.e-8
    learning_rate: 1.e-3
    batch_norm: False
    net_dropout: 0
    batch_size: 128
    embedding_dim: 4
    epochs: 1
    shuffle: True
    seed: 2019
    monitor: 'AUC'
    monitor_mode: 'max'
```

--------------------------------

### Run DCN Experiment

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/DCN/DCN_torch/README.md

Execute the DCN model experiment using the provided script. Specify the experiment ID and the GPU to use.

```python
python run_expid.py --expid DCN_test --gpu 0
```

--------------------------------

### Dataset Configuration (Compact)

Source: https://github.com/reczoo/fuxictr/blob/main/docs/configurations.md

Defines dataset parameters including data paths, format, and feature columns. Multiple features can be grouped with shared settings for compactness.

```yaml
# dataset_config.yaml
taubao_tiny: # dataset_id
    data_root: ../data/
    data_format: csv
    train_data: ../data/tiny_data/train_sample.csv
    valid_data: ../data/tiny_data/valid_sample.csv
    test_data: ../data/tiny_data/test_sample.csv
    min_categr_count: 1
    feature_cols:
        - {name: ["userid","adgroup_id","pid","cate_id","campaign_id","customer","brand","cms_segid",
                  "cms_group_id","final_gender_code","age_level","pvalue_level","shopping_level","occupation"], 
                  active: True, dtype: str, type: categorical}
    label_col: {name: clk, dtype: float}
```

--------------------------------

### Dataset Configuration (Expanded)

Source: https://github.com/reczoo/fuxictr/blob/main/docs/configurations.md

Defines dataset parameters with each feature column explicitly listed. This format is more verbose but can be clearer for individual feature settings.

```yaml
taobao_tiny:
    data_root: ../data/
    data_format: csv
    train_data: ../data/tiny_data/train_sample.csv
    valid_data: ../data/tiny_data/valid_sample.csv
    test_data: ../data/tiny_data/test_sample.csv
    min_categr_count: 1
    feature_cols:
        [{name: "userid", active: True, dtype: str, type: categorical},
         {name: "adgroup_id", active: True, dtype: str, type: categorical},
         {name: "pid", active: True, dtype: str, type: categorical},
         {name: "cate_id", active: True, dtype: str, type: categorical},
         {name: "campaign_id", active: True, dtype: str, type: categorical},
         {name: "customer", active: True, dtype: str, type: categorical},
         {name: "brand", active: True, dtype: str, type: categorical},
         {name: "cms_segid", active: True, dtype: str, type: categorical},
         {name: "cms_group_id", active: True, dtype: str, type: categorical},
         {name: "final_gender_code", active: True, dtype: str, type: categorical},
         {name: "age_level", active: True, dtype: str, type: categorical},
         {name: "pvalue_level", active: True, dtype: str, type: categorical},
         {name: "shopping_level", active: True, dtype: str, type: categorical},
         {name: "occupation", active: True, dtype: str, type: categorical}]
    label_col: {name: clk, dtype: float}
```

--------------------------------

### Python Requirements for QNN

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/QNN/README.md

Specifies the required Python version, PyTorch version, and fuxictr library version for running the QNN model.

```python
python: 3.8
pytorch: 1.10
fuxictr: 2.0.1
```

--------------------------------

### Layer-wise Activation Specification

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/FinalNet/README.md

Shows how to define layer-specific activation functions within network blocks. Activations can be a single string for all layers or a list for individual layer assignments.

```text
block1_hidden_activations: ["relu",  "leakyrelu", "sigmoid"]
```

```text
block2_hidden_activations: ["relu", "leakyrelu", "sigmoid"]
```

--------------------------------

### FinalMLP Requirements

Source: https://github.com/reczoo/fuxictr/blob/main/model_zoo/FinalMLP/README.md

Specifies the required versions for Python, PyTorch, and fuxictr to run the FinalMLP model.

```python
python: 3.6/3.7
pytorch: 1.0/1.11
fuxictr: 2.0.1
```

--------------------------------

### DeepFM Model Configuration

Source: https://github.com/reczoo/fuxictr/blob/main/docs/configurations.md

This configuration block defines parameters for training a DeepFM model. It includes settings for model architecture, dataset, training process, and optimization.

```yaml
DeepFM_test:
    model_root: '../checkpoints/'
    workers: 3
    verbose: 1
    patience: 2
    pickle_feature_encoder: True
    use_hdf5: True
    save_best_only: True
    every_x_epochs: 1
    debug: False
    partition_block_size: -1
    model: DeepFM
    dataset_id: taobao_tiny_data
    loss: 'binary_crossentropy'
    metrics: ['logloss', 'AUC']
    task: binary_classification
    optimizer: adam
    hidden_units: [64, 32]
    hidden_activations: relu
    net_regularizer: 0
    embedding_regularizer: 1.e-8
    learning_rate: 1.e-3
    batch_norm: False
    net_dropout: 0
    batch_size: 128
    embedding_dim: 4
    epochs: 1
    shuffle: True
    seed: 2019
    monitor: 'AUC'
    monitor_mode: 'max'
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.