A preprocessing layer which hashes and bins categorical features.

This layer transforms categorical inputs to hashed output. It element-wise converts a ints or strings to ints in a fixed range. The stable hash function uses tensorflow::ops::Fingerprint to produce the same output consistently across all platforms.

This layer uses FarmHash64 by default, which provides a consistent hashed output across different platforms and is stable across invocations, regardless of device and context, by mixing the input bits thoroughly.

If you want to obfuscate the hashed output, you can also pass a random salt argument in the constructor. In that case, the layer will use the SipHash64 hash function, with the salt value serving as additional input to the hash function.

Note: This layer internally uses TensorFlow. It cannot be used as part of the compiled computation graph of a model with any backend other than TensorFlow. It can however be used with any backend when running eagerly. It can also always be used as part of an input preprocessing pipeline with any backend (outside the model itself), which is how we recommend to use this layer.

Note: This layer is safe to use inside a tf.data pipeline (independently of which backend you're using).

Example (FarmHash64)

layer <- layer_hashing(num_bins = 3)
inp <- c('A', 'B', 'C', 'D', 'E') |> array(dim = c(5, 1))
layer(inp)

## tf.Tensor(
## [[1]
##  [0]
##  [1]
##  [1]
##  [2]], shape=(5, 1), dtype=int64)

Example (FarmHash64) with a mask value

layer <- layer_hashing(num_bins=3, mask_value='')
inp <- c('A', 'B', '', 'C', 'D') |> array(dim = c(5, 1))
layer(inp)

## tf.Tensor(
## [[1]
##  [1]
##  [0]
##  [2]
##  [2]], shape=(5, 1), dtype=int64)

Example (SipHash64)

layer <- layer_hashing(num_bins=3, salt=c(133, 137))
inp <- c('A', 'B', 'C', 'D', 'E') |> array(dim = c(5, 1))
layer(inp)

## tf.Tensor(
## [[1]
##  [2]
##  [1]
##  [0]
##  [2]], shape=(5, 1), dtype=int64)

Example (Siphash64 with a single integer, same as salt=[133, 133])

layer <- layer_hashing(num_bins=3, salt=133)
inp <- c('A', 'B', 'C', 'D', 'E') |> array(dim = c(5, 1))
layer(inp)

## tf.Tensor(
## [[0]
##  [0]
##  [2]
##  [1]
##  [0]], shape=(5, 1), dtype=int64)

Usage

layer_hashing(
  object,
  num_bins,
  mask_value = NULL,
  salt = NULL,
  output_mode = "int",
  sparse = FALSE,
  ...
)

Arguments

object

Object to compose the layer with. A tensor, array, or sequential model.

num_bins

Number of hash bins. Note that this includes the mask_value bin, so the effective number of bins is (num_bins - 1) if mask_value is set.

mask_value

A value that represents masked inputs, which are mapped to index 0. NULL means no mask term will be added and the hashing will start at index 0. Defaults to NULL.

salt

A single unsigned integer or NULL. If passed, the hash function used will be SipHash64, with these values used as an additional input (known as a "salt" in cryptography). These should be non-zero. If NULL, uses the FarmHash64 hash function. It also supports list of 2 unsigned integer numbers, see reference paper for details. Defaults to NULL.

output_mode

Specification for the output of the layer. Values can be "int", "one_hot", "multi_hot", or "count" configuring the layer as follows:

"int": Return the integer bin indices directly.
"one_hot": Encodes each individual element in the input into an array the same size as num_bins, containing a 1 at the input's bin index. If the last dimension is size 1, will encode on that dimension. If the last dimension is not size 1, will append a new dimension for the encoded output.
"multi_hot": Encodes each sample in the input into a single array the same size as num_bins, containing a 1 for each bin index index present in the sample. Treats the last dimension as the sample dimension, if input shape is (..., sample_length), output shape will be (..., num_tokens).
"count": As "multi_hot", but the int array contains a count of the number of times the bin index appeared in the sample. Defaults to "int".

sparse

Boolean. Only applicable to "one_hot", "multi_hot", and "count" output modes. Only supported with TensorFlow backend. If TRUE, returns a SparseTensor instead of a dense Tensor. Defaults to FALSE.

...

Keyword arguments to construct a layer.

Value

The return value depends on the value provided for the first argument. If object is:

a keras_model_sequential(), then the layer is added to the sequential model (which is modified in place). To enable piping, the sequential model is also returned, invisibly.
a keras_input(), then the output tensor from calling layer(input) is returned.
NULL or missing, then a Layer instance is returned.

Input Shape

A single string, a list of strings, or an int32 or int64 tensor of shape (batch_size, ...,).

Output Shape

An int32 tensor of shape (batch_size, ...).

Reference

SipHash with salt

Other preprocessing layers:
layer_category_encoding()
layer_center_crop()
layer_discretization()
layer_feature_space()
layer_hashed_crossing()
layer_integer_lookup()
layer_mel_spectrogram()
layer_normalization()
layer_random_brightness()
layer_random_contrast()
layer_random_crop()
layer_random_flip()
layer_random_rotation()
layer_random_translation()
layer_random_zoom()
layer_rescaling()
layer_resizing()
layer_string_lookup()
layer_text_vectorization()

Other layers:
Layer()
layer_activation()
layer_activation_elu()
layer_activation_leaky_relu()
layer_activation_parametric_relu()
layer_activation_relu()
layer_activation_softmax()
layer_activity_regularization()
layer_add()
layer_additive_attention()
layer_alpha_dropout()
layer_attention()
layer_average()
layer_average_pooling_1d()
layer_average_pooling_2d()
layer_average_pooling_3d()
layer_batch_normalization()
layer_bidirectional()
layer_category_encoding()
layer_center_crop()
layer_concatenate()
layer_conv_1d()
layer_conv_1d_transpose()
layer_conv_2d()
layer_conv_2d_transpose()
layer_conv_3d()
layer_conv_3d_transpose()
layer_conv_lstm_1d()
layer_conv_lstm_2d()
layer_conv_lstm_3d()
layer_cropping_1d()
layer_cropping_2d()
layer_cropping_3d()
layer_dense()
layer_depthwise_conv_1d()
layer_depthwise_conv_2d()
layer_discretization()
layer_dot()
layer_dropout()
layer_einsum_dense()
layer_embedding()
layer_feature_space()
layer_flatten()
layer_flax_module_wrapper()
layer_gaussian_dropout()
layer_gaussian_noise()
layer_global_average_pooling_1d()
layer_global_average_pooling_2d()
layer_global_average_pooling_3d()
layer_global_max_pooling_1d()
layer_global_max_pooling_2d()
layer_global_max_pooling_3d()
layer_group_normalization()
layer_group_query_attention()
layer_gru()
layer_hashed_crossing()
layer_identity()
layer_integer_lookup()
layer_jax_model_wrapper()
layer_lambda()
layer_layer_normalization()
layer_lstm()
layer_masking()
layer_max_pooling_1d()
layer_max_pooling_2d()
layer_max_pooling_3d()
layer_maximum()
layer_mel_spectrogram()
layer_minimum()
layer_multi_head_attention()
layer_multiply()
layer_normalization()
layer_permute()
layer_random_brightness()
layer_random_contrast()
layer_random_crop()
layer_random_flip()
layer_random_rotation()
layer_random_translation()
layer_random_zoom()
layer_repeat_vector()
layer_rescaling()
layer_reshape()
layer_resizing()
layer_rnn()
layer_separable_conv_1d()
layer_separable_conv_2d()
layer_simple_rnn()
layer_spatial_dropout_1d()
layer_spatial_dropout_2d()
layer_spatial_dropout_3d()
layer_spectral_normalization()
layer_string_lookup()
layer_subtract()
layer_text_vectorization()
layer_tfsm()
layer_time_distributed()
layer_torch_module_wrapper()
layer_unit_normalization()
layer_upsampling_1d()
layer_upsampling_2d()
layer_upsampling_3d()
layer_zero_padding_1d()
layer_zero_padding_2d()
layer_zero_padding_3d()
rnn_cell_gru()
rnn_cell_lstm()
rnn_cell_simple()
rnn_cells_stack()