View Source Scholar.Preprocessing.OneHotEncoder (Scholar v0.3.1)
Implements encoder that converts integer value (substitute of categorical data in tensors) into 0-1 vector. The index of 1 in the vector is aranged in sorted manner. This means that for x < y => one_index(x) < one_index(y).
Currently the module supports only 1D tensors.
Summary
Functions
Creates mapping from values into one-hot vectors.
Apply encoding on the provided tensor directly. It's equivalent to fit/2 and then transform/2 on the same data.
Encode labels as a one-hot numeric tensor. All values provided to transform/2 must be seen
in fit/2 function, otherwise an error occurs.
Functions
Creates mapping from values into one-hot vectors.
Options
:num_classes(pos_integer/0) - Required. Number of classes to be encoded.
Examples
iex> t = Nx.tensor([3, 2, 4, 56, 2, 4, 2])
iex> Scholar.Preprocessing.OneHotEncoder.fit(t, num_classes: 4)
%Scholar.Preprocessing.OneHotEncoder{
encoder: %Scholar.Preprocessing.OrdinalEncoder{
encoding_tensor: Nx.tensor(
[
[0, 2],
[1, 3],
[2, 4],
[3, 56]
]
)
},
one_hot: Nx.tensor(
[
[1, 0, 0, 0],
[0, 1, 0, 0],
[0, 0, 1, 0],
[0, 0, 0, 1]
], type: :u8
)
}Apply encoding on the provided tensor directly. It's equivalent to fit/2 and then transform/2 on the same data.
Examples
iex> t = Nx.tensor([3, 2, 4, 56, 2, 4, 2])
iex> Scholar.Preprocessing.OneHotEncoder.fit_transform(t, num_classes: 4)
#Nx.Tensor<
u8[7][4]
[
[0, 1, 0, 0],
[1, 0, 0, 0],
[0, 0, 1, 0],
[0, 0, 0, 1],
[1, 0, 0, 0],
[0, 0, 1, 0],
[1, 0, 0, 0]
]
>Encode labels as a one-hot numeric tensor. All values provided to transform/2 must be seen
in fit/2 function, otherwise an error occurs.
Examples
iex> t = Nx.tensor([3, 2, 4, 56, 2, 4, 2])
iex> enoder = Scholar.Preprocessing.OneHotEncoder.fit(t, num_classes: 4)
iex> Scholar.Preprocessing.OneHotEncoder.transform(enoder, t)
#Nx.Tensor<
u8[7][4]
[
[0, 1, 0, 0],
[1, 0, 0, 0],
[0, 0, 1, 0],
[0, 0, 0, 1],
[1, 0, 0, 0],
[0, 0, 1, 0],
[1, 0, 0, 0]
]
>
iex> t = Nx.tensor([3, 2, 4, 56, 2, 4, 2])
iex> enoder = Scholar.Preprocessing.OneHotEncoder.fit(t, num_classes: 4)
iex> new_tensor = Nx.tensor([2, 3, 4, 3, 4, 56, 2])
iex> Scholar.Preprocessing.OneHotEncoder.transform(enoder, new_tensor)
#Nx.Tensor<
u8[7][4]
[
[1, 0, 0, 0],
[0, 1, 0, 0],
[0, 0, 1, 0],
[0, 1, 0, 0],
[0, 0, 1, 0],
[0, 0, 0, 1],
[1, 0, 0, 0]
]
>