tf2.0-CSV文件的生成及使用方法

tf2.0-CSV文件的生成及使用方法

from sklearn.datasets import fetch_california_housing
housing = fetch_california_housing()

数据形式为每八个特征对应一条房价

之后对数据进行数据集分割以及归一化处理：

from sklearn.model_selection import train_test_split
train_x_all, test_x, train_y_all, test_y = train_test_split(housing.data, housing.target, random_state=5)
train_x, valid_x, train_y, valid_y = train_test_split(train_x_all, train_y_all, random_state=8)
print(train_x.shape, test_x.shape, valid_x.shape)
print(train_y.shape, test_y.shape, valid_y.shape)

from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(train_x)
x_valid_scaled = scaler.transform(valid_x)
x_test_scaled = scaler.transform(test_x)

output_dir = "generate_csv"
if not os.path.exists(output_dir):
  os.mkdir(output_dir)
# data：希望转化为csv的数据；name_prefix：train/valid/test
# header：一条数据中特征名称组成的字符串； n_parts：将data存储到n_parts个csv文件中
def save_to_csv(output_dir, data, name_prefix, header=None, n_parts=10):
  path_format = os.path.join(output_dir, "{}_{:02d}.csv")
  filenames = []
"""对于每个part，利用索引定位待存储数据， row_indices即为当前part中存在的数据的索引数组"""
  for file_idx, row_indices in enumerate(np.array_split(np.arange(len(data)), n_parts)):
    part_csv = path_format.format(name_prefix, file_idx)
    filenames.append(part_csv)
    with open(part_csv, 'wt', encoding='utf-8') as f:
      if header is not None:
        f.write(header + '\n')
"""将每条数据的每维信息变成字符串形式保存在文件中"""
      for row_index in row_indices:
        f.write(','.join([repr(col) for col in data[row_index]]))
        f.write('\n')
  return filenames
""" np.c_:将两矩阵按列拼接"""
train_data = np.c_[x_train_scaled, train_y]
valid_data = np.c_[x_valid_scaled, valid_y]
test_data = np.c_[x_test_scaled, test_y]
header_cols = housing.feature_names+["price"]
header_str = ','.join(header_cols)


train_filenames = save_to_csv(output_dir, train_data, "train", header_str, 20)
valid_filenames = save_to_csv(output_dir, valid_data, 'valid', header_str, 10)
test_filenames = save_to_csv(output_dir, test_data, 'test', header_str, 10)

csv文件的读取需要经过以下步骤：
1.将文件名（filenames）处理成一个dataset
2.读取文件，生成所有filename对应的dataset，并将各dataset合并
3.将csv文件中文本内容解析出来：
tf.io.decode_csv(str, record_defaults)

将文件名（filenames）处理成一个dataset:

# 生成文件名数据集, tf.data.Dataset.list_files将文件名处理成一个dataset
filename_dataset = tf.data.Dataset.list_files(train_filenames)

读取文件，生成所有filename对应的dataset，并将各dataset合并:

# interleave()作用于dataset并将其合并
dataset = filename_dataset.interleave(
# 将filename中数据按行读取形成一个dataset并将各filename对应的dataset合并为一个大的dataset, skip用于取消表头
    lambda filename: tf.data.TextLineDataset(filename).skip(1),
#并行读取文件个数
    cycle_length = 5
)
#可以查看dataset中前15行数据
for line in dataset.take(15):
  print(line)

将csv文件中文本内容解析出来：

def parse_csv_line(line, n_fields):
  defs = [tf.constant(np.nan)] * n_fields
  parsed_fields = tf.io.decode_csv(line, defs)
#将值变成向量：tf.stack()  
  x = tf.stack(parsed_fields[:-1])
  y = tf.stack(parsed_fields[-1:])
  return x, y

合并以上三步的整体过程：

def csv_reader_dataset(filenames, n_readers=5, batch_size=32, n_parse_threads=5, shuffle_buffer_size=10000):
  dataset = tf.dataset.list_files(filenames)
  dataset = dataset.repeat()
  dataset = dataset.interleave(
      lambda filename: tf.data.TextLineDataset(filename),
      cycle_length = n_readers)
  dataset.shuffle(shuffle_buffer_size)
  dataset = dataset.map(parse_csv_line, num_parallel_calls=n_parse_threads)
  dataset = dataset.batch(batch_size)
  return dataset

生成datasets:

train_set = csv_reader_dataset(train_filenames)
valid_set = csv_reader_dataset(valid_filenames)
test_set = csv_reader_dataset(test_filenames)

model = keras.models.Sequential()
model.add(keras.layers.Dense(30,input_shape=[8], activation="relu"))
model.add(keras.layers.Dense(1))
model.compile(loss="mse", optimizer="sgd")
callbacks=[keras.callbacks.EarlyStopping(patience=5, min_delta=1e-3)]
model.fit(train_set, validation_data=valid_set, epochs = 100, steps_per_epoch = 11600//32, validation_steps = 3870//32, callbacks=callbacks)

进行一下测试：

model.evaluate(test_set, steps=5160//batch_size)