IAfLow

This library help to create models with identifiers, checkpoints, logs and metadata automatically, in order to make the training process more efficient and traceable.

Requirements

• Tensorflow 2.5 or higher.
• notify-function 1.5.0 or higher.

Usage

This library has a maser class called IAFlow, that has functions to management model creation and training, the most important are:

• add_model: to add a new model to the internal structure of maker models to build later to training.
• train: to train a model with a internal callbacks to save metrics with Tensorboard, save the model with checkpoints and send notification to discord channel, email or telegram (all this is optional).
• show_models: to show the models that are in the internal structure of maker models.
• delete_model: to delete a model from the internal structure of maker models.

There are more methods to complete a CRUD for models and datasets. You can check it once you create a instance of IAFlow. For now, you could check a brief description of most important methods below.

Constructor

from iaflow import IAFlow

def custom_builder(input_shape):
  model = tf.keras.models.Sequential([
    tf.keras.layers.Input(input_shape),
    tf.keras.layers.Dense(units=512, activation='relu'),
    tf.keras.layers.Dense(units=512, activation='relu'),
    tf.keras.layers.Dense(units=1, activation='sigmoid')
  ])
  return model

params_notifier = { # Parameters for notifier, see documentation https://pypi.org/project/notify-function/#description
  'title': 'Training update',
  'webhook_url': os.environ.get('WEBHOOK_URL'),
  'frequency_epoch': 20 # This will send a notification every 20 epochs, by default it is every epoch
}

ia_maker = IAFlow(
  models_folder='./models', # Folder to save the models
  params_notifier=params_notifier, # Notifier to send notification to discord channel, email or telegram (all this is optional)
  builder_function=custom_builder # Function to build the model, you can change this when call `train` method
)

add_dataset

This function add a new dataset to the internal structure of maker models to build later to training. The dataset could be an instance of tf.data.Dataset, list of Tensors of whatever that could be passes to the fit method of tf.keras.Model.

ia_maker.add_dataset(
  name='dataset_1', # Name of the dataset
  epochs=10, # Number of epochs to train the model with this dataset
  batch_size=32, # Batch size to train the model with this dataset
  shuffle_buffer=512, # Buffer size to shuffle the dataset when training with this dataset
  train_ds=train_ds, # Dataset to train the model, you can change this when call `train` method
  val_ds=val_ds # Dataset to validate the model, you can change this when call `train` method
)

So far, the shuffle buffer is only applied to datasets that are an instance of tf.data.Dataset.

add_model method

The build method is used to create a model in a defined structure of folder. As below:

`models_folder`
└─ `model_name`
   └─ `run_id`
      ├─ logs                               
      │  ├─ train
      │  ├─ validation
      ├─ `model_name`_`model_params`_checkpoint.h5
      └─ `model_name`_params.json

model_1_data = ia_maker.add_model(
  model_name='model_1', # Name of the model
  run_id='run_1', # Run id of the model, must be unique by model name
  model_params={ 'input_shape': (2, 1) }, # Parameters for builder function
  load_model_params={}, # Parameters for model loading, see documentation of tf.keras.models.load_model
  compile_params={ # Parameters for model compilation, see documentation tf.keras.Models.compile
    'optimizer': 'adam', 'loss': 'mse',
    'metrics': ['accuracy']
  },
)

Note: If your model use custom SubClass for custom Layers you must send the parameter load_model_params with the parameter custom_objects with the custom class. see documentation of tf.keras.models.load_model

Return

This method return a dictionary with information of the model

train method

The train method is used to train a model. This doen't return anything. Inside the method the model will be built, compiled and finally begin to train.

Parameters

ia_maker.train(model_1_data, batch=32, epochs=5)

If you loaded a trained model you should send a dictionary with key model_name and run_id that belong to the model desired to load and the initial epoch, also you can send the same dictionary returned when you called add_model method.

If you want use notify-function lib you must send the parameter params_notifier. To see how to use and what do this library you can see the documentation of notify-function. The message will have the epoch number and the values of each metric that you defined for your model.

If you use notify-function and you specify email maybe this will add a delay to the training process if your time per step is so fast. To avoid this you can use another methods more faster like discord webhooks or telegram message or instead use the key frequency_epoch on the params_notifier reduce the rate of notifications.

show_models method

This function show all models that you have added to the IAFlow by a print.

delete_model method

This function delete a model from the IAFlow. You must send a dictionary with key model_name and run_id that belong to the model desired to delete. If you send the parameter delete_folder it will also delete the folder of the model.

FAQs

If you have any question find a bug or feedback, please contact me with a email to [email protected]

Made with ❤️ by Enmanuel Magallanes