官方教程詳解

Part 1 –資料集無法下載

在命令列模式下執行 python premade_estimator.py時會遇到報錯，錯誤原因是資料集無法下載。在iris_data.py這個檔案下，我們可以看到通過tf.keras.utils.get_file這個函式將資料集從”http://download.tensorflow.org/data/iris_training.csv“中拷貝過來。但由於無法訪問該網址，因此會出現報錯。解決辦法：通過其他途徑將iris的資料集（150條資料）下載，然後拆分為訓練集（120條資料）和驗證集（30條資料）。iris_data.py檔案的程式改動如下：

def maybe_download():
    #train_path = tf.keras.utils.get_file(TRAIN_URL.split('/')[-1], TRAIN_URL)
    #test_path = tf.keras.utils.get_file(TEST_URL.split('/')[-1], TEST_URL)
    train_path='/Users/stuyu/Desktop/Iris_dataset/iris_train.csv'
    test_path='/Users/stuyu/Desktop/Iris_dataset/iris_test.csv'
    return
 
 train_path, test_path

Part 2 –命令列解析模組

在程式的開端premade_estimator.py檔案下，我們可以看到：

parser = argparse.ArgumentParser()
parser.add_argument('--batch_size', default=100, type=int, help='batch size')
parser.add_argument('--train_steps', default=1000, type=int,
                    help='number of training steps'
 
)

這一段程式碼是命令列解析程式，在命令列執行該python檔案時，我們可以通過加入–batch_size 和 –train_steps 來改變batch_size和train_step。
詳情請見https://www.cnblogs.com/yymn/p/8056487.html

Part 3 –特徵編碼

當程式繼續執行時，會遇到報錯

Traceback (most recent call last):
  File "/Users/stuyu/Documents/PythonCode/tf-slim/models/samples/core/get_started/premade_estimator.py", line 118, in <module>
    tf.app.run(main)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 126, in run
    _sys.exit(main(argv))
  File "/Users/stuyu/Documents/PythonCode/tf-slim/models/samples/core/get_started/premade_estimator.py", line 71, in main
    steps=args.train_steps)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 363, in train
    loss = self._train_model(input_fn, hooks, saving_listeners)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 843, in _train_model
    return self._train_model_default(input_fn, hooks, saving_listeners)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 856, in _train_model_default
    features, labels, model_fn_lib.ModeKeys.TRAIN, self.config)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 831, in _call_model_fn
    model_fn_results = self._model_fn(features=features, **kwargs)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/canned/dnn.py", line 347, in _model_fn
    config=config)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/canned/dnn.py", line 190, in _dnn_model_fn
    logits=logits)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/canned/head.py", line 771, in create_estimator_spec
    features=features, mode=mode, logits=logits, labels=labels)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/canned/head.py", line 682, in create_loss
    label_ids = self._label_ids(labels)
  File "/Users/stuyu/anaconda3/lib/python3.6/site-packages/tensorflow/python/estimator/canned/head.py", line 665, in _label_ids
    format(labels.dtype))
ValueError: Labels dtype should be integer. Instead got <dtype: 'string'>.

報錯的最後一行提示我們Label應該為整數型，而不應該為字串。檢視資料，我們可以看到label基本是’Setosa’, ‘Versicolor’, ‘Virginica’。需要轉變成整數型。因此我們要用到sklearn庫裡的preprocessing.LabelEncoder。

#iris_data.py

from sklearn.preprocessing import LabelEncoder
...
def load_data(y_name='Species'):
    ...
    labelEncoder=LabelEncoder()
    train_y=labelEncoder.fit_transform(train_y)
    test_y=labelEncoder.fit_transform(test_y)

    return (train_x, train_y), (test_x, test_y)

由於資料集中是’Setosa’, ‘Versicolor’, ‘Virginica’依次排列的，故分別對應0，1，2。恰好和下面程式對應上了（內心一陣竊喜）。

#premade_estimator.py

# Generate predictions from the model
    expected = ['Setosa', 'Versicolor', 'Virginica']

此時，再次執行命令便可以成功運行了。