(filename, batch_size, num_epochs)
| 409 | |
| 410 | |
| 411 | def build_model_input(filename, batch_size, num_epochs): |
| 412 | |
| 413 | def parse_parquet(value): |
| 414 | tf.logging.info('Parsing {}'.format(filename)) |
| 415 | labels = value.pop(LABEL_COLUMN[0]) |
| 416 | dense_feature = [value[name] for name in CONTINUOUS_COLUMNS] |
| 417 | |
| 418 | sparse_feature = [value[name] for name in CATEGORICAL_COLUMNS] |
| 419 | return dense_feature, sparse_feature, labels |
| 420 | |
| 421 | '''Work Queue Feature''' |
| 422 | if args.workqueue and not args.tf: |
| 423 | from tensorflow.python.ops.work_queue import WorkQueue |
| 424 | work_queue = WorkQueue([filename], num_epochs=num_epochs) |
| 425 | # For multiple files： |
| 426 | # work_queue = WorkQueue([filename, filename1,filename2,filename3]) |
| 427 | files = work_queue.input_dataset() |
| 428 | else: |
| 429 | files = filename |
| 430 | |
| 431 | from tensorflow.python.data.experimental.ops import parquet_dataset_ops |
| 432 | |
| 433 | dataset = parquet_dataset_ops.ParquetDataset(files, batch_size=batch_size) |
| 434 | dataset = dataset.map(parse_parquet, num_parallel_calls=28) |
| 435 | dataset = dataset.prefetch(2) |
| 436 | return dataset |
| 437 | |
| 438 | |
| 439 | def model_fn(strategy, sparse_feature, dense_feature): |
no test coverage detected