Add support for databases which contain more than one sample in one file

bob/learn/tensorflow/script/db_to_tfrecords.py

@@ -73,6 +73,9 @@ The config files should have the following objects totally:
   # this is False, the loaded samples from a file are iterated over and each
   # of them is saved as an independent feature.
   one_file_one_sample = True
+
+  # Converts the read data to this format. [default: float32]
+  data_type = "float32"
 """
 
 from __future__ import absolute_import
@@ -96,7 +99,7 @@ def _int64_feature(value):
 
 
 def write_a_sample(writer, data, label):
-    feature = {'train/data': _bytes_feature(data.astype('float32').tostring()),
+    feature = {'train/data': _bytes_feature(data.tostring()),
                'train/label': _int64_feature(label)}
 
     example = tf.train.Example(features=tf.train.Features(feature=feature))
@@ -127,6 +130,7 @@ def main(argv=None):
     data_extension = getattr(config, 'data_extension', '.hdf5')
     shuffle = getattr(config, 'shuffle', False)
     one_file_one_sample = getattr(config, 'one_file_one_sample', True)
+    data_type = getattr(config, 'data_type', "float32")
 
     create_directories_safe(output_dir)
     if not isinstance(groups, (list, tuple)):
@@ -143,7 +147,7 @@ def main(argv=None):
                 logger.info('Processing file %d out of %d', i + 1, n_files)
 
                 path = f.make_path(data_dir, data_extension)
-                data = reader(path)
+                data = reader(path).astype(data_type)
                 label = file_to_label(f)
 
                 if one_file_one_sample: