如何记录从TensorFlow到CSV文件的结果

Question

我有一个在TensorFlow上运行的CNN模型，并希望保存准确性，损失，F1，精度和回忆值，因为我还想保存这些图和混淆矩阵（您可以将这些剧情保存到CSV吗？）。 如何将每个模型运行到CSV或文本文件中保存此数据？

Answer 1

尝试使用tf.keras.callbacks.csvlogger：

import tensorflow as tf
import pandas as pd

model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(1, input_dim=40))
model.add(tf.keras.layers.Dense(1, 'sigmoid'))

adam_opt = tf.keras.optimizers.Adam(0.1)
model.compile(loss='bce', optimizer=adam_opt, metrics=[tf.keras.metrics.BinaryAccuracy(name="binary_accuracy", dtype=None), 
                                                       tf.keras.metrics.Recall()])

train_x = tf.random.normal((50, 40))
train_y = tf.random.uniform((50, 1), maxval=2, dtype=tf.int32)
val_x = tf.random.normal((50, 40))
val_y = tf.random.uniform((50, 1), maxval=2, dtype=tf.int32)

csv_logger = tf.keras.callbacks.CSVLogger('metrics.csv')
history = model.fit(train_x, train_y, epochs=5, validation_data=(val_x, val_y), callbacks=[csv_logger])

df = pd.read_csv('/content/metrics.csv')
print(df.to_markdown())

Epoch 1/5
2/2 [==============================] - 2s 563ms/step - loss: 0.7918 - binary_accuracy: 0.4400 - recall: 0.4583 - val_loss: 0.7283 - val_binary_accuracy: 0.4200 - val_recall: 0.4815
Epoch 2/5
2/2 [==============================] - 0s 62ms/step - loss: 0.6793 - binary_accuracy: 0.5400 - recall: 0.5417 - val_loss: 0.7093 - val_binary_accuracy: 0.4200 - val_recall: 0.2593
Epoch 3/5
2/2 [==============================] - 0s 92ms/step - loss: 0.6647 - binary_accuracy: 0.6200 - recall: 0.3750 - val_loss: 0.7138 - val_binary_accuracy: 0.4400 - val_recall: 0.2222
Epoch 4/5
2/2 [==============================] - 0s 68ms/step - loss: 0.6369 - binary_accuracy: 0.6200 - recall: 0.3750 - val_loss: 0.7340 - val_binary_accuracy: 0.4400 - val_recall: 0.3704
Epoch 5/5
2/2 [==============================] - 0s 69ms/step - loss: 0.5869 - binary_accuracy: 0.6800 - recall: 0.5417 - val_loss: 0.7975 - val_binary_accuracy: 0.4800 - val_recall: 0.4444

	epoch	binary_accuracy	loss	recall	val_binary_accuracy	val_loss	val_recall
0	0	0.44	0.791773	0.458333	0.42	0.728296	0.481481
1	1	0.54	0.67928	0.541667	0.42	0.709347	0.259259
2	2	0.62	0.664661	0.375	0.44	0.713829	0.222222
3	3	0.62	0.636919	0.375	0.44	0.734033	0.37037
4	4	0.68	0.586907	0.541667	0.48	0.797542	0.444444

After training,您可以轻松地使用CSV文件进行绘图。