Fix local dataset generation in TFRecord example notebook (aws#682)

This commit also makes the dataset generation and model creation cells re-runnable, and adds a wait to the batch transform step.

Author: apacker

advanced_functionality/working_with_tfrecords/working-with-tfrecords.ipynb

@@ -90,12 +90,13 @@
  " return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))\n",
  "\n",
  "# create the tfrecord dataset dir\n",
- "os.mkdir(tfrecord_root)\n",
+ "if not os.path.isdir(tfrecord_root):\n",
+ " os.mkdir(tfrecord_root)\n",
  "\n",
  "for input_file, output_file in [(test_csv_file,test_tfrecord_file), (train_csv_file,train_tfrecord_file)]:\n",
  " # create the output file\n",
  " open(tfrecord_root + output_file, 'a').close()\n",
- " with tf.python_io.TFRecordWriter(output_file) as writer:\n",
+ " with tf.python_io.TFRecordWriter(tfrecord_root + output_file) as writer:\n",
  " with open(csv_root + input_file,'r') as f:\n",
  " f.readline() # skip first line\n",
  " for line in f:\n",
@@ -105,7 +106,7 @@
  " 'petal_length': _floatlist_feature(line.split(',')[2]),\n",
  " 'petal_width': _floatlist_feature(line.split(',')[3]),\n",
  " }\n",
- " if file == train_csv_file:\n",
+ " if f == train_csv_file:\n",
  " feature['label'] = _int64list_feature(int(line.split(',')[4].rstrip()))\n",
  " example = tf.train.Example(\n",
  " features=tf.train.Features(\n",
@@ -266,10 +267,11 @@
  "outputs": [],
  "source": [
  "from sagemaker.tensorflow.serving import Model\n",
+ "from sagemaker.utils import name_from_base\n",
  "\n",
  "client = boto3.client('sagemaker')\n",
  "\n",
- "model_name = 'tfrecord-to-tfserving'\n",
+ "model_name = name_from_base('tfrecord-to-tfserving')\n",
  "\n",
  "transform_container = {\n",
  " \"Image\": transformer_repository_uri\n",
@@ -324,7 +326,8 @@
  " sagemaker_session=sess,\n",
  ")\n",
  "transformer.transform(data = input_data_path,\n",
- " split_type = 'TFRecord')"
+ " split_type = 'TFRecord')\n",
+ "transformer.wait()"
  ]
  },
  {