Блог

Я пытаюсь реализовать последовательную модель регрессии с Keras, и я получаю очень странные результаты. Мой код приведен ниже. Я использую набор данных tf.data в качестве своего входного набора данных. Сначала мои потери снижаются, но затем начинают колебаться. Есть ли причина, по которой это происходит?

     #initialize model
    model = keras.models.Sequential()
    
    model.add(Dense(1 (len(feature_names)-1), input_shape=((len(feature_names)-1),)))
    model.add(LeakyReLU())
    model.add(Dropout(0.5))
    
    model.add(Dense(10))
    model.add(LeakyReLU())
    
    #output layer, 1 unit
    model.add(Dense(1))
    
    loss = 'mean_squared_error'
    optimizer = tf.keras.optimizers.Nadam(learning_rate=0.0001)
    
    #compile the model
    m = tf.keras.metrics.RootMeanSquaredError()
    model.compile(optimizer=optimizer, loss=loss, metrics=[m,'mae'])
    
#    early_stop = keras.callbacks.EarlyStopping(monitor='val_loss', patience=4)
    tb = TensorBoard(histogram_freq=1)
    
    ##try learning rates, different noise, dropout, using only nasadem
    
    #run the model
    history = model.fit(x=final_train_dataset,epochs=count,steps_per_epoch=steps_per_epoch,validation_data=final_valid_dataset,callbacks=[tb])
  

 5468/5468 [==============================] - 19s 4ms/step - loss: 19.8461 - root_mean_squared_error: 4.4549 - mae: 3.1814 - val_loss: 13.2963 - val_root_mean_squared_error: 3.6464 - val_mae: 3.0513
Epoch 2/25
   1/5468 [..............................] - ETA: 0s - loss: 12.9226 - root_mean_squared_error: 3.5948 - mae: 2.63732020-09-03 17:02:51.641109: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 216632 of 1000000
2020-09-03 17:03:01.641093: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 432841 of 1000000
2020-09-03 17:03:11.641090: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 644957 of 1000000
2020-09-03 17:03:14.202555: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 9ms/step - loss: 4.5396 - root_mean_squared_error: 2.1306 - mae: 1.5395 - val_loss: 9.2400 - val_root_mean_squared_error: 3.0397 - val_mae: 2.3317
Epoch 3/25
   1/5468 [..............................] - ETA: 0s - loss: 2.1542 - root_mean_squared_error: 1.4677 - mae: 1.14152020-09-03 17:03:43.456237: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 209810 of 1000000
2020-09-03 17:03:53.456238: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 430811 of 1000000
2020-09-03 17:04:03.456243: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 649754 of 1000000
2020-09-03 17:04:05.717317: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 51s 9ms/step - loss: 3.5548 - root_mean_squared_error: 1.8854 - mae: 1.3483 - val_loss: 12.9111 - val_root_mean_squared_error: 3.5932 - val_mae: 3.1018
Epoch 4/25
   1/5468 [..............................] - ETA: 0s - loss: 3.0977 - root_mean_squared_error: 1.7600 - mae: 1.38832020-09-03 17:04:34.262132: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 220010 of 1000000
2020-09-03 17:04:44.262123: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 440231 of 1000000
2020-09-03 17:04:54.262103: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 658897 of 1000000
2020-09-03 17:04:56.431462: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 9ms/step - loss: 1.9917 - root_mean_squared_error: 1.4113 - mae: 1.0305 - val_loss: 18.9676 - val_root_mean_squared_error: 4.3552 - val_mae: 3.4140
Epoch 5/25
   1/5468 [..............................] - ETA: 0s - loss: 2.0044 - root_mean_squared_error: 1.4158 - mae: 0.98262020-09-03 17:05:25.768521: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 209530 of 1000000
2020-09-03 17:05:35.768491: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 415149 of 1000000
2020-09-03 17:05:45.768498: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 619178 of 1000000
2020-09-03 17:05:49.626056: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 10ms/step - loss: 3.1466 - root_mean_squared_error: 1.7739 - mae: 1.2621 - val_loss: 9.5821 - val_root_mean_squared_error: 3.0955 - val_mae: 2.3618
Epoch 6/25
   1/5468 [..............................] - ETA: 0s - loss: 2.2696 - root_mean_squared_error: 1.5065 - mae: 1.13252020-09-03 17:06:18.050735: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 216085 of 1000000
2020-09-03 17:06:28.050765: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 433756 of 1000000
2020-09-03 17:06:38.050744: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 647129 of 1000000
2020-09-03 17:06:40.516620: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 51s 9ms/step - loss: 6.5886 - root_mean_squared_error: 2.5668 - mae: 1.9638 - val_loss: 12.8445 - val_root_mean_squared_error: 3.5839 - val_mae: 2.6729
Epoch 7/25
   1/5468 [..............................] - ETA: 0s - loss: 6.2392 - root_mean_squared_error: 2.4978 - mae: 1.95852020-09-03 17:07:08.786525: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 211271 of 1000000
2020-09-03 17:07:18.786501: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 413948 of 1000000
2020-09-03 17:07:28.786525: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 628456 of 1000000
2020-09-03 17:07:32.091402: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 51s 9ms/step - loss: 8.6716 - root_mean_squared_error: 2.9448 - mae: 2.2906 - val_loss: 8.4462 - val_root_mean_squared_error: 2.9062 - val_mae: 2.0556
Epoch 8/25
   1/5468 [..............................] - ETA: 0s - loss: 7.3663 - root_mean_squared_error: 2.7141 - mae: 2.09252020-09-03 17:07:59.862839: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 196030 of 1000000
2020-09-03 17:08:09.862769: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 396211 of 1000000
2020-09-03 17:08:19.862786: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 596521 of 1000000
2020-09-03 17:08:24.550654: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 53s 10ms/step - loss: 4.6825 - root_mean_squared_error: 2.1639 - mae: 1.5158 - val_loss: 3.3543 - val_root_mean_squared_error: 1.8315 - val_mae: 1.4609
Epoch 9/25
   1/5468 [..............................] - ETA: 0s - loss: 3.3096 - root_mean_squared_error: 1.8192 - mae: 1.34092020-09-03 17:08:52.847230: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 213665 of 1000000
2020-09-03 17:09:02.847221: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 420633 of 1000000
2020-09-03 17:09:12.847262: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 625401 of 1000000
2020-09-03 17:09:16.325973: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 53s 10ms/step - loss: 7.0318 - root_mean_squared_error: 2.6518 - mae: 2.0913 - val_loss: 5.9001 - val_root_mean_squared_error: 2.4290 - val_mae: 1.7496
Epoch 10/25
   1/5468 [..............................] - ETA: 0s - loss: 5.4777 - root_mean_squared_error: 2.3405 - mae: 1.84602020-09-03 17:09:45.541587: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 211860 of 1000000
2020-09-03 17:09:55.541592: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 425901 of 1000000
2020-09-03 17:10:05.541639: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 638506 of 1000000
2020-09-03 17:10:08.444475: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 51s 9ms/step - loss: 5.5155 - root_mean_squared_error: 2.3485 - mae: 1.6887 - val_loss: 7.0059 - val_root_mean_squared_error: 2.6469 - val_mae: 1.9491
Epoch 11/25
   1/5468 [..............................] - ETA: 0s - loss: 4.7595 - root_mean_squared_error: 2.1816 - mae: 1.60422020-09-03 17:10:36.978236: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 220628 of 1000000
2020-09-03 17:10:46.978223: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 439429 of 1000000
2020-09-03 17:10:56.978219: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 658497 of 1000000
2020-09-03 17:10:58.887525: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 50s 9ms/step - loss: 7.7528 - root_mean_squared_error: 2.7844 - mae: 2.1798 - val_loss: 7.4304 - val_root_mean_squared_error: 2.7259 - val_mae: 2.1477
Epoch 12/25
   1/5468 [..............................] - ETA: 0s - loss: 5.9227 - root_mean_squared_error: 2.4337 - mae: 1.86982020-09-03 17:11:26.748628: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 215455 of 1000000
2020-09-03 17:11:36.748645: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 432008 of 1000000
2020-09-03 17:11:46.748652: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 648981 of 1000000
2020-09-03 17:11:49.102683: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 9ms/step - loss: 5.3337 - root_mean_squared_error: 2.3095 - mae: 1.7406 - val_loss: 16.7965 - val_root_mean_squared_error: 4.0984 - val_mae: 2.9269
Epoch 13/25
   1/5468 [..............................] - ETA: 0s - loss: 3.7040 - root_mean_squared_error: 1.9246 - mae: 1.49812020-09-03 17:12:18.282279: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 220812 of 1000000
2020-09-03 17:12:28.282279: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 446392 of 1000000
2020-09-03 17:12:38.282271: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 665649 of 1000000
2020-09-03 17:12:39.861825: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 9ms/step - loss: 8.0562 - root_mean_squared_error: 2.8384 - mae: 2.1867 - val_loss: 16.5570 - val_root_mean_squared_error: 4.0690 - val_mae: 3.2326
Epoch 14/25
   1/5468 [..............................] - ETA: 0s - loss: 7.7569 - root_mean_squared_error: 2.7851 - mae: 2.19592020-09-03 17:13:10.059481: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 218698 of 1000000
2020-09-03 17:13:20.059483: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 442427 of 1000000
2020-09-03 17:13:30.059488: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 667466 of 1000000
2020-09-03 17:13:31.590296: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 50s 9ms/step - loss: 5.5334 - root_mean_squared_error: 2.3523 - mae: 1.7780 - val_loss: 2.7601 - val_root_mean_squared_error: 1.6614 - val_mae: 1.3854
Epoch 15/25
   1/5468 [..............................] - ETA: 0s - loss: 5.4189 - root_mean_squared_error: 2.3279 - mae: 1.71692020-09-03 17:14:00.199430: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 221723 of 1000000
2020-09-03 17:14:10.199432: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 439283 of 1000000
2020-09-03 17:14:20.199441: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 656318 of 1000000
2020-09-03 17:14:22.153380: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 50s 9ms/step - loss: 7.9082 - root_mean_squared_error: 2.8121 - mae: 2.2185 - val_loss: 5.2582 - val_root_mean_squared_error: 2.2931 - val_mae: 1.8866
Epoch 16/25
   1/5468 [..............................] - ETA: 0s - loss: 8.4431 - root_mean_squared_error: 2.9057 - mae: 2.26912020-09-03 17:14:49.720142: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 206675 of 1000000
2020-09-03 17:14:59.720147: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 421484 of 1000000
2020-09-03 17:15:09.720163: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 621983 of 1000000
2020-09-03 17:15:13.378677: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 53s 10ms/step - loss: 8.1283 - root_mean_squared_error: 2.8510 - mae: 2.2215 - val_loss: 5.8047 - val_root_mean_squared_error: 2.4093 - val_mae: 1.7678
Epoch 17/25
   1/5468 [..............................] - ETA: 0s - loss: 6.9673 - root_mean_squared_error: 2.6396 - mae: 2.06012020-09-03 17:15:42.974799: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 216994 of 1000000
2020-09-03 17:15:52.974823: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 431703 of 1000000
2020-09-03 17:16:02.974787: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 647963 of 1000000
2020-09-03 17:16:05.286592: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 10ms/step - loss: 5.6694 - root_mean_squared_error: 2.3811 - mae: 1.8346 - val_loss: 4.0211 - val_root_mean_squared_error: 2.0053 - val_mae: 1.4926
Epoch 18/25
   1/5468 [..............................] - ETA: 0s - loss: 4.3882 - root_mean_squared_error: 2.0948 - mae: 1.69472020-09-03 17:16:35.245666: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 219541 of 1000000
2020-09-03 17:16:45.245695: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 434849 of 1000000
2020-09-03 17:16:55.245697: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 657326 of 1000000
2020-09-03 17:16:57.154813: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 51s 9ms/step - loss: 7.6286 - root_mean_squared_error: 2.7620 - mae: 2.1647 - val_loss: 8.8990 - val_root_mean_squared_error: 2.9831 - val_mae: 2.1807
Epoch 19/25
   1/5468 [..............................] - ETA: 0s - loss: 6.3365 - root_mean_squared_error: 2.5172 - mae: 1.98792020-09-03 17:17:25.767055: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 210541 of 1000000
2020-09-03 17:17:35.767069: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 428168 of 1000000
2020-09-03 17:17:45.767050: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 648394 of 1000000
2020-09-03 17:17:48.163076: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 9ms/step - loss: 5.6925 - root_mean_squared_error: 2.3859 - mae: 1.8557 - val_loss: 7.0036 - val_root_mean_squared_error: 2.6464 - val_mae: 2.0192
Epoch 20/25
   1/5468 [..............................] - ETA: 0s - loss: 5.9111 - root_mean_squared_error: 2.4313 - mae: 1.86942020-09-03 17:18:17.564046: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 218969 of 1000000
2020-09-03 17:18:27.564032: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 438295 of 1000000
2020-09-03 17:18:37.564054: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 655413 of 1000000
2020-09-03 17:18:39.599428: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 10ms/step - loss: 6.8373 - root_mean_squared_error: 2.6148 - mae: 1.9742 - val_loss: 4.4934 - val_root_mean_squared_error: 2.1198 - val_mae: 1.5544
Epoch 21/25
  20/5468 [..............................] - ETA: 16s - loss: 6.7229 - root_mean_squared_error: 2.5929 - mae: 1.95182020-09-03 17:19:09.851911: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 210717 of 1000000
2020-09-03 17:19:19.851916: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 414573 of 1000000
2020-09-03 17:19:29.851935: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 631533 of 1000000
2020-09-03 17:19:33.007503: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 9ms/step - loss: 5.7252 - root_mean_squared_error: 2.3927 - mae: 1.7921 - val_loss: 19.9770 - val_root_mean_squared_error: 4.4696 - val_mae: 3.9483
Epoch 22/25
  20/5468 [..............................] - ETA: 14s - loss: 5.5166 - root_mean_squared_error: 2.3488 - mae: 1.75092020-09-03 17:20:01.793350: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 206965 of 1000000
2020-09-03 17:20:11.793374: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 417262 of 1000000
2020-09-03 17:20:21.793379: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 618303 of 1000000
2020-09-03 17:20:25.882237: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 53s 10ms/step - loss: 7.3727 - root_mean_squared_error: 2.7153 - mae: 2.1631 - val_loss: 3.7680 - val_root_mean_squared_error: 1.9411 - val_mae: 1.4858
Epoch 23/25
  22/5468 [..............................] - ETA: 13s - loss: 6.6480 - root_mean_squared_error: 2.5784 - mae: 2.03622020-09-03 17:20:55.131738: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 197145 of 1000000
2020-09-03 17:21:05.131721: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 411294 of 1000000
2020-09-03 17:21:15.131712: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 624435 of 1000000
2020-09-03 17:21:18.642404: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 52s 9ms/step - loss: 5.5347 - root_mean_squared_error: 2.3526 - mae: 1.7962 - val_loss: 8.6975 - val_root_mean_squared_error: 2.9492 - val_mae: 2.3858
Epoch 24/25
  23/5468 [..............................] - ETA: 13s - loss: 5.2042 - root_mean_squared_error: 2.2813 - mae: 1.72872020-09-03 17:21:46.952287: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 217458 of 1000000
2020-09-03 17:21:56.952289: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 421999 of 1000000
2020-09-03 17:22:06.952280: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 614871 of 1000000
2020-09-03 17:22:11.265809: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 53s 10ms/step - loss: 2.9234 - root_mean_squared_error: 1.7098 - mae: 1.2129 - val_loss: 8.1747 - val_root_mean_squared_error: 2.8591 - val_mae: 2.3386
Epoch 25/25
  23/5468 [..............................] - ETA: 12s - loss: 2.7464 - root_mean_squared_error: 1.6572 - mae: 1.18782020-09-03 17:22:39.998529: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 210233 of 1000000
2020-09-03 17:22:49.998541: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 420717 of 1000000
2020-09-03 17:22:59.998540: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:184] Filling up shuffle buffer (this may take a while): 631855 of 1000000
2020-09-03 17:23:03.105398: I tensorflow/core/kernels/data/shuffle_dataset_op.cc:233] Shuffle buffer filled.
5468/5468 [==============================] - 51s 9ms/step - loss: 3.2256 - root_mean_squared_error: 1.7960 - mae: 1.2697 - val_loss: 8.8438 - val_root_mean_squared_error: 2.9738 - val_mae: 2.2013
  

Я взглянул на гистограммы тензорной доски и получил странную гистограмму для одного из слоев, но я не знаю, как ее интерпретировать.

Любая помощь в расшифровке того, что происходит, была бы очень признательна!

Гистограммы, которые вы видите, представляют собой распределения веса двух плотных слоев.
Каждый плотный слой имеет веса (w1, w2, .., wn) и смещение (b) . Веса называются ядрами в Tensorflow.

Гистограммы представляют ядра и смещения двух плотных слоев.

Как их интерпретировать?

Поскольку это 3d-гистограмма, сначала она содержит три оси.

x-axis — Фактическое значение, которое принимают ядра и смещения. т.Е. Значение b, w1, w2 и т. Д
y-axis — Номер эпохи.
z-axis — Количество весов

Мы возьмем dense_1/bias_0 для нашего примера. В этом
x-axis : от -0.8 до 1.6
Интерпретация — Все значения в векторе смещения слоя dense_1 находятся в диапазоне от -0.8 до 1.6.

y-axis : 0 — 22
Интерпретация — Гистограмма показывает значения для каждой эпохи от 0 до 22.

z-axis :
Интерпретация — Это показывает, сколько смещений имеет значение, представленное осью x . т.е.. Если большее количество отклонений имеет значение, равное нулю, вы получите самый высокий пик (наибольшее значение z) при x = 0.

Итак, из гистограммы для dense_1 / bias_0 вы можете интерпретировать это.
При epoch = 0 все отклонения имели значение 0. Поскольку вы получаете пик при x = 0 и не получаете значения z для других значений x.

При epoch = 22 ваш вектор смещения не содержит никакого значения между 0.0 и 0.4.

Почему используются эти гистограммы?
Эти гистограммы используются, чтобы увидеть, действительно ли модель обучается или нет. Если гистограмма остается неизменной для каждой эпохи, то ваша модель не обучается должным образом и возникает некоторая ошибка.

Полезные ссылки:

1. Руководство по тензорной доске
2. Использование тензорной доски для отладки