Keras2.xのCallbackを自作するときのメモ

Callbackでオーバーライドする時にメソッドの引数の中身が不明だったので、ログ出力してみた。
オーバーライドできるメソッドはgithubを参照すること。

from keras.callbacks import Callback

class MyCallback(Callback) :

    def __init__(self) :

    def on_train_begin(self, logs=None):
        print("on_train_begin")
        print("model", self.model)
        print("params", self.params)
        print("logs", logs)
        print("=" * 20)

    def on_train_end(self, logs=None):
        print("on_train_end")
        print("logs", logs)
        print("=" * 20)

    def on_batch_begin(self, batch, logs = None) :
        print("on_batch_begin")
        print("batch", batch)
        print("logs", logs)
        print("=" * 20)

    def on_batch_end(self, batch, logs = None) :
        print("on_batch_end")
        print("batch", batch)
        print("logs", logs)
        print("=" * 20)

    def on_epoch_begin(self, epoch, logs=None):
        print("on_epoch_begin")
        print("epoch", epoch)
        print("logs", logs)
        print("=" * 20)

    def on_epoch_end(self, epoch, logs = None) :
        print("on_epoch_end")
        print("epoch", epoch)
        print("logs", logs)
        print("=" * 20)

# Modelの構築は省略して、fitの箇所だけ書いてます。
model.fit(data, label_one_hot, epochs=3, batch_size=50, validation_split=0.0, callbacks=[MyCallback()], verbose=0)

ログ

on_train_begin
model <keras.engine.sequential.Sequential object at 0x7f114e96b400>
params {'batch_size': 50, 'epochs': 3, 'steps': None, 'samples': 90, 'verbose': 0, 'do_validation': False, 'metrics': ['loss', 'acc']}
logs {}
====================
on_epoch_begin
epoch 0
logs {}
====================
on_batch_begin
batch 0
logs {'batch': 0, 'size': 50}
====================
on_batch_end
batch 0
logs {'batch': 0, 'size': 50, 'loss': 1.1077641, 'acc': 0.42}
====================
on_batch_begin
batch 1
logs {'batch': 1, 'size': 40}
====================
on_batch_end
batch 1
logs {'batch': 1, 'size': 40, 'loss': 1.0845511, 'acc': 0.6}
====================
on_epoch_end
epoch 0
logs {'loss': 1.097447223133511, 'acc': 0.5000000033113692}
====================
on_epoch_begin
epoch 1
logs {}
====================
on_batch_begin
batch 0
logs {'batch': 0, 'size': 50}
====================
on_batch_end
batch 0
logs {'batch': 0, 'size': 50, 'loss': 1.0686755, 'acc': 0.72}
====================
on_batch_begin
batch 1
logs {'batch': 1, 'size': 40}
====================
on_batch_end
batch 1
logs {'batch': 1, 'size': 40, 'loss': 1.0953418, 'acc': 0.375}
====================
on_epoch_end
epoch 1
logs {'loss': 1.0805271996392145, 'acc': 0.5666666825612386}
====================
on_epoch_begin
epoch 2
logs {}
====================
on_batch_begin
batch 0
logs {'batch': 0, 'size': 50}
====================
on_batch_end
batch 0
logs {'batch': 0, 'size': 50, 'loss': 1.0497301, 'acc': 0.76}
====================
on_batch_begin
batch 1
logs {'batch': 1, 'size': 40}
====================
on_batch_end
batch 1
logs {'batch': 1, 'size': 40, 'loss': 1.1001902, 'acc': 0.275}
====================
on_epoch_end
epoch 2
logs {'loss': 1.0721567736731634, 'acc': 0.5444444417953491}
====================
on_train_end
logs {}
====================

まとめ

タイミング	メソッド名	引数	内容
訓練開始時	on_train_begin	logs	{}
訓練終了時	on_train_end	logs	{}
epoch開始時	on_epoch_begin	epoch	int 0開始
		logs	{}
epoch終了時	on_epoch_end	epoch	int 0開始
		logs	{'loss': 1.093177596728007, 'acc': 0.36666665805710685}
1batchの訓練開始時	on_batch_begin	batch	int 0開始
		logs	{'batch': 0, 'size': 50}
1batchの訓練終了時	on_batch_end	batch	int 0開始
		logs	{'batch': 0, 'size': 50, 'loss': 1.1000787, 'acc': 0.48}