tf.contrib.framework.load_variable_slot_initializer

tf.contrib.framework.load_variable_slot_initializer(
    ckpt_path,
    old_tensor_name,
    primary_partition_info,
    new_row_vocab_size,
    new_col_vocab_size,
    old_row_vocab_file=None,
    new_row_vocab_file=None,
    old_col_vocab_file=None,
    new_col_vocab_file=None,
    num_row_oov_buckets=0,
    num_col_oov_buckets=0,
    initializer=None,
    max_rows_in_memory=-1
)

Defined in tensorflow/contrib/framework/python/ops/checkpoint_ops.py.

Loads pre-trained multi-class slots for linear models from checkpoint.

Wrapper around load_and_remap_matrix_initializer() specialized for loading multi-class slots (such as optimizer accumulators) and remapping them according to the provided vocab files. See docs for load_and_remap_matrix_initializer() for more details. Takes in a variable_scope._PartitionInfo representing the slot's primary Variable's partitioning. This is necessary since accumulator Variable creation ignores primary scoping and partitioning information.

Args:

• ckpt_path: Path to the TensorFlow checkpoint (version 2, TensorBundle) from which the old matrix Tensor will be loaded.
• old_tensor_name: Name of the 2-D Tensor to load from checkpoint.
• primary_partition_info: A variable_scope._PartitionInfo containing this slot's primary Variable's partitioning information. This is used to calculate the offset and override the partition_info passed to the call to _initialize.
• new_row_vocab_size: int specifying the number of entries in new_row_vocab_file. If no row remapping is needed (no row vocab provided), this should be equal to the number of rows to load from the old matrix (which can theoretically be smaller than the number of rows in the old matrix).
• new_col_vocab_size: int specifying the number of entries in new_col_vocab_file. If no column remapping is needed (no column vocab provided), this should be equal to the number of columns in the old matrix.
• old_row_vocab_file: A scalar Tensor of type string containing the path to the old row vocabulary file. Can be None, which represents no remapping on the row axis.
• new_row_vocab_file: A scalar Tensor of type string containing the path to the new row vocabulary file. Can be None, which represents no remapping on the row axis.
• old_col_vocab_file: A scalar Tensor of type string containing the path to the old column vocabulary file. Can be None, which represents no remapping on the column axis.
• new_col_vocab_file: A scalar Tensor of type string containing the path to the new column vocabulary file. Can be None, which represents no remapping on the column axis.
• num_row_oov_buckets: int specifying the number of out-of-vocabulary rows to append. Must be >= 0.
• num_col_oov_buckets: int specifying the number of out-of-vocabulary columns to append. Must be >= 0.
• initializer: Initializer function to initialize missing values. Accepts a 1-D tensor as the arg to specify the shape of the returned tensor. If None, defaults to using zeros_initializer().
• max_rows_in_memory: int specifying the maximum number of rows to load from the checkpoint at once. If less than or equal to 0, the entire matrix will be loaded into memory. Setting this arg trades increased disk reads for lower memory usage.

Returns:

A variable initializer function that should be used to initialize a (potentially partitioned) Variable whose complete shape is [new_row_vocab_size + num_row_oov_buckets, new_col_vocab_size + num_col_oov_buckets].

Raises:

• TypeError: If initializer is specified but not callable.