Metadata-Version: 2.1
Name: neurostage
Version: 0.2
Summary: A framework for managing deep learning projects
Home-page: https://github.com/catalina-delgado/NeuroStage
Author: Catalina Delgado
Author-email: catalina08delgado@gmail.com
License: UNKNOWN
Keywords: training,deepLearning,framework
Platform: UNKNOWN
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.6
Description-Content-Type: text/markdown
Requires-Dist: numpy<2.0.0,>=1.21.0
Requires-Dist: tensorflow<3.0.0,>=2.10.0
Requires-Dist: opencv-python<5.0.0,>=4.5.0
Requires-Dist: tensorboard<3.0.0,>=2.10.0

# NeuroStage
"NeuroStage is a framework that allows users to create and manage deep learning projects in a structured and modular way, adapted for TensorFlow. It includes integration with tools like Tensorboard, enabling users to efficiently track and improve their models."

# Purpose
NeuroStage was born from the idea of automatically generating projects, with a specific focus on building deep learning models using TensorFlow. It is a tool designed for new users who need a standard structure without having to worry about organizing the project from scratch.

# Índice

1. [Design](#Design)
2. [Features](#Features)
3. [Installation](#installation) 
4. [Usage Flow](#usage-flow)

# Design
It is designed as a layer-based pattern (for building modules, architectures, and training) which is excellent for organizing a TensorFlow framework for deep learning testing. This modular approach facilitates integration with TensorBoard and promotes scalability. 

## Modules

**Layers** Define base layers here (e.g., convolutional, attention, etc.) that can be used in models. These layers form the building blocks for your deep learning models.

**Models** Combine the layers to create specific architectures for evaluation. This module allows you to design and implement various deep learning models by reusing and combining different layers.

**Training** Conduct experiments with specific configurations, logging metrics, parameters, and artifacts. This module focuses on the training process, helping you to configure and run training sessions, and track the performance and results.

# Features

| Feature                  | DeepTrain                                              |
|--------------------------|--------------------------------------------------------|
| Model Management         | Allows customization for versioning and saving models. |
| Test Automation          | Executes each training session in series as defined by the training module. |                                                       |
| Tool Compatibility       | TensorFlow, TensorBoard, OpenCV, Numpy                                            |
| Open Source              | MIT License                                            |
| Flexibility              | Preconfigured but flexible, define rules and processes as per your case |
| Collaboration            | Avalable                                               |


## Project Structure
```
my_project/
│
├── config.py             # Project configuration file
├── utils.py              # General utilities file
├── functions.py          # Training functions file
├── imports.py            # Library imports file
├── experiments/          # Folder for experiments
└── src/                  # Main source code folder
    ├── layers/           # Folder for implementing custom layers
    │   └── layer_a.py    # Example content
    │   └── layer_b.py 
    ├── models/           # Folder for defining models
    │   └── model_a.py    # Example content
    │   └── model_b.py 
    └── training/         # Folder for compiling and starting training
        └── train_a.py    # Example content
        └── train_b.py
```
# Installation
To install **NeuroStage**, simply run the following command:
``` 
pip install neurostage
```
For more detailed information, visit the project page on Github:
[NeuroStage](https://github.com/catalina-delgado/NeuroStage)

## stage
To confirm that the installation was successful, run the following command:
```
stage
```
You should see the following output:
```
usage: stage [-h] [--list] [--help_train] {startproject,run} ...

NeuroStage: A framework for testing and training deep learning models.

positional arguments:
  {startproject,run}
    startproject      Start a New Project
    run               Run the current project

options:
  -h, --help          show this help message and exit
  --list              List the available models
  --help_train        Show usage examples
```
If you see this message, the installation has been completed successfully and the `stage` command is ready to use.

# Usage-Flow
## Start a new project
To start a new project, use the following command. You can replace `my_project` with your desired project name:
```
stage startproject my_project
```
## create a new layer
File: src/layers/layer_custom.py
```python
from imports import tf

class CustomLayer(tf.keras.layers.Layer):
    def __init__(self, units, **kwargs):
        super().__init__(**kwargs)
        self.units = units
        self.dense = tf.keras.layers.Dense(units)

    def call(self, inputs):
        return self.dense(inputs)

```
## create a new model
File: src/models/model_custom.py
```python
from imports import tf
from src.layers.layer_custom import CustomLayer

class ModelCustom():
    def __init__(self): 
        super(Model, self).__init__() 
        self.layer = CustomLayer(64)


    def build_model(self, input):
        x = self.layer(input)
        x = self.layer(x)
        x = self.layer(x)

        model = tf.keras.Model(inputs=input, outputs=x)

        return model
```
## Create a training runner
File: src/training/train_custom.py
```python
from functions import NeuroStage
from imports import tf, np
from src.models.model import Model

class TrainModel(NeuroStage):
    def __init__(self, batch_size=32, epochs=4, model_name='', models=None):
        super().__init__()

        self.BATCH_SIZE = batch_size
        self.EPHOCS = epochs
        self.MODEL_NAME = model_name

        input = tf.keras.Input(shape=(256, 256, 1))  
        self.optimizer = tf.keras.optimizers.SGD(learning_rate=1e-3, momentum=0.95)
        self.architecture = Model()
        print(models)
        self.model = self.architecture.build_model(input)
        self.model.compile(optimizer=self.optimizer,
                         loss='binary_crossentropy',
                         metrics=['accuracy'])

    def train(self):
        X_train = np.random.rand(100, 256, 256, 1)
        y_train = np.random.randint(0, 2, 100) 
        X_val = np.random.rand(20, 256, 256, 1) 
        y_val = np.random.randint(0, 2, 20)

        self.init_fit(self.model, X_train, y_train, X_val, y_val, self.EPHOCS, self.BATCH_SIZE, self.MODEL_NAME)
```
## Execution
```
stage run --batch_size 32 --epochs 10 --model_name my_model
```


