TensorFlow Datasets and Estimato

• TextLineDataset从text中读取lines
  • Returns:
    A list of Tensor objects. Has the same type as record_defaults.
  • 默认情况，TextLineDataset yields every line of each file

#filter以#开头的注释部分
dataset = dataset.flat_map(
    lambda filename: (
        tf.data.TextLineDataset(filename)
        .skip(1)
        .filter(lambda line: tf.not_equal(tf.substr(line, 0, 1), "#"))))

• TFRecordDataset TFRecord中读取
• FixedLengthRecordDataset 二进制文件读取固定大小的record

Dataset

• tf.decode_csv
  • Returns:
    A list of Tensor objects. Has the same type as record_defaults.
    Each tensor will have the same shape as records.

Estimator.jpg

• input_fn provides the estimator with input data

feature_columns = [tf.feature_column.numeric_column(k) for k in feature_names] #features_names为特征名称的list

classifier = tf.estimator.DNNClassifier(
   feature_columns=feature_columns, # The input features to our model
   hidden_units=[10, 10], # 2层 10个neurons
   n_classes=3,
   model_dir=PATH)