def make_source_dataset(self, index, num_hosts):
"""See base class."""
data = self.selection
client = contrib_cloud.BigtableClient(data.project, data.instance)
table = client.table(data.table)
ds = table.parallel_scan_prefix(
data.prefix, columns=[(data.column_family, data.column_qualifier)])
# The Bigtable datasets will have the shape (row_key, data)
ds_data = ds.map(lambda index, data: data)
if self.is_training:
ds_data = ds_data.repeat()
return ds_data
def __init__(self, project_name, instance_name, table_name):
"""Constructor.
Args:
project_name: string name of GCP project having table.
instance_name: string name of CBT instance in project.
table_name: string name of CBT table in instance.
"""
self.btspec = BigtableSpec(project_name, instance_name, table_name)
self.bt_table = bigtable.Client(
self.btspec.project,
admin=True).instance(self.btspec.instance).table(self.btspec.table)
self.tf_table = contrib_cloud.BigtableClient(
self.btspec.project, self.btspec.instance).table(self.btspec.table)
