11class AblationStudy (object ):
2- def __init__ (self , training_dataset_name , training_dataset_version ,
3- label_name , ** kwargs ):
2+ """The `AblationStudy` object is the entry point to define an ablation
3+ study with maggy. This object can subsequently be passed as an argument
4+ when the experiment is launched with `experiment.lagom()`.
5+
6+ Sample usage:
7+
8+ >>> from maggy.ablation import AblationStudy
9+ >>> ablation_study = AblationStudy('titanic_train_dataset',
10+ >>> label_name='survived')
11+
12+ Define your study by including layers and features, which should be
13+ ablated:
14+
15+ >>> ablation_study.features.include('pclass', 'fare')
16+ >>> ablation_study.model.layers.include('my_dense_two',
17+ >>> 'my_dense_three')
18+
19+ You can also add a layer group using a list:
20+
21+ >>> ablation_study.model.layers.include_groups(['my_dense_two',
22+ >>> 'my_dense_four'])
23+
24+ Or add a layer group using a prefix:
25+
26+ >>> ablation_study.model.layers.include_groups(prefix='my_dense')
27+
28+ Next you should define a base model function using the layer and feature
29+ names you previously specified:
30+
31+ >>> # you only need to add the `name` parameter to layer initializers
32+ >>> def base_model_generator():
33+ >>> model = tf.keras.Sequential()
34+ >>> model.add(tf.keras.layers.Dense(64, activation='relu'))
35+ >>> model.add(tf.keras.layers.Dense(..., name='my_dense_two', ...)
36+ >>> model.add(tf.keras.layers.Dense(32, activation='relu'))
37+ >>> model.add(tf.keras.layers.Dense(..., name='my_dense_sigmoid', ...)
38+ >>> # output layer
39+ >>> model.add(tf.keras.layers.Dense(1, activation='linear'))
40+ >>> return model
41+
42+ Make sure to include the generator function in the study:
43+
44+ >>> ablation_study.model.set_base_model_generator(base_model_generator)
45+
46+ Last but not least you can define your actual training function:
47+
48+ >>> from maggy import experiment
49+ >>> from maggy.callbacks import KerasBatchEnd
50+
51+ >>> def training_function(dataset_function, model_function, reporter):
52+ >>> import tensorflow as tf
53+ >>> epochs = 5
54+ >>> batch_size = 10
55+ >>> tf_dataset = dataset_function(epochs, batch_size)
56+ >>> model = model_function()
57+ >>> model.compile(optimizer=tf.train.AdamOptimizer(0.001),
58+ >>> loss='binary_crossentropy',
59+ >>> metrics=['accuracy'])
60+ >>> ### Maggy REPORTER
61+ >>> callbacks = [KerasBatchEnd(reporter, metric='acc')]
62+ >>> history = model.fit(tf_dataset, epochs=5, steps_per_epoch=30)
63+ >>> return float(history.history['acc'][-1])
64+
65+ Lagom the experiment:
66+
67+ >>> result = experiment.lagom(map_fun=training_function,
68+ >>> experiment_type='ablation',
69+ >>> ablation_study=ablation_study,
70+ >>> ablator='loco',
71+ >>> name='Titanic-LOCO',
72+ >>> hb_interval=5)
73+ """
74+
75+ def __init__ (
76+ self , training_dataset_name , training_dataset_version , label_name ,
77+ ** kwargs ):
78+ """Initializes the ablation study.
79+
80+ :param training_dataset_name: Name of the training dataset in the
81+ featurestore.
82+ :type training_dataset_name: str
83+ :param training_dataset_version: Version of the training dataset to be
84+ used.
85+ :type training_dataset_version: int
86+ :param label_name: Name of the target prediction label.
87+ :type label_name: str
88+ """
489 self .features = Features ()
590 self .model = Model ()
691 self .hops_training_dataset_name = training_dataset_name
@@ -11,8 +96,9 @@ def __init__(self, training_dataset_name, training_dataset_version,
1196 def to_dict (self ):
1297 """
1398 Returns the ablation study configuration as a Python dictionary.
14- :return: A dictionary with ablation study configuration parameters as keys
15- (i.e. 'training_dataset_name', 'included_features', etc.)
99+
100+ :return: A dictionary with ablation study configuration parameters as
101+ keys (i.e. 'training_dataset_name', 'included_features', etc.)
16102 :rtype: dict
17103 """
18104 ablation_dict = {
0 commit comments