rndtools 3.0.5

R&D tools

Installation

pip install rndtools

Mini-framework for keras model training

When you call train_model function then this framework will do few useful things: 1. it creates dirs when it do not exist 1. it automatically save architecture into atrchitecture.json, 1. plot model graph 1. save python source code of get_model_function and training_function, 1. after each epoch draw loss and accuray chart, 1. after each epoch save csv file with learning history. 1. saves some meta information about model, for example: 1. processing time 1. best train and test loss 1. date of model creation

What you should do:

1. Implement load_data function.

2. Implement function that returns compiled keras model. Function should not have any parameters. Example:

   def get_model():
     model = Sequential()
     model.add(Dense(12, input_dim=8, init='uniform', activation='relu'))
     model.add(Dense(8, init='uniform', activation='relu'))
     model.add(Dense(1, init='uniform', activation='sigmoid'))
   
     model.compile(
       optimizer=Adam(),
       loss='binary_crossentropy',
       metrics=['accuracy']
     )
   
    return model

3. Implement function that trains model. Function should return model history. Example: ``` def train(data, model, model_folder, callbacks=None): if callbacks is None: callbacks = []

   history = model.fit(data.X, data.Y, nb_epoch=150, batch_size=10, callbacks=callbacks)
   
   return history

   ``Pay attention to callbacks parameter. There are some extra callbacks that you should add to model callbacks. Also note that as indataparameter function pass whatload_data` function returns.

Example:

>>> import rndtools as rnd
>>> rnd.train.train_model(
    model_dir,
    get_model_function=get_model,
    training_function=train,
    loading_data_function=load_data
)

Model path: /home/rd/notebooks/documents-detector/damian/models/in_the_wild/unet_mrz/7

------------------------------
Creating dirs...
------------------------------
------------------------------
Creating and compiling model...
------------------------------
------------------------------
Saving architecture...
------------------------------
------------------------------
Plotting model...
------------------------------
------------------------------
Saving model source code...
------------------------------
------------------------------
Loading data...
------------------------------
------------------------------
Instantiating callbacks...
------------------------------
------------------------------
Training model...
------------------------------
Epoch 1/1000

Finished!

Dataset In Parts Generator

Sometimes there is so many data that it is problem to store it in memory. Then you can use divide your dataset into parts DatasetInPartsGenerator that will load this parts in turn, so you will have only part of dataset in memory.

Pipeline

Pipeline consists of steps. Each step is a tuple (name of step, step class). Example pipeline:

pipeline = Pipeline(
    (
        'grayscale',
        Grayscale()
    ),
    (
        'threshold',
        Threshold()
    ),
    (
        'blur',
        Blur(
            sigma=1.5
        )
    ),
    (
        'watershed',
        Watershed(
            min_distance=5,
            threshold_rel=0.1
        )
    ),
    show_progressbar=True
)

To create your own step just inherit from Step and implement transform method:

from rndtools.pipeline import Step

class CustomStep(Step):
    def transform(self, params):
        pass