mirror of
https://github.com/ml-explore/mlx.git
synced 2025-08-20 10:27:41 +08:00
implemented batchnorm layer
This commit is contained in:
parent
22fee5a383
commit
2b617b63bd
@ -1,5 +1,7 @@
|
||||
# Copyright © 2023 Apple Inc.
|
||||
|
||||
from typing import Tuple
|
||||
|
||||
import mlx.core as mx
|
||||
from mlx.nn.layers.base import Module
|
||||
|
||||
@ -97,7 +99,7 @@ class GroupNorm(Module):
|
||||
where :math:`\gamma` and :math:`\beta` are learned per feature dimension
|
||||
parameters initialized at 1 and 0 respectively. However, the mean and
|
||||
variance are computed over the spatial dimensions and each group of
|
||||
features. In particular, the input is split into num_groups across the
|
||||
features. In particular, the input is split into num_groups accross the
|
||||
feature dimension.
|
||||
|
||||
The feature dimension is assumed to be the last dimension and the dimensions
|
||||
@ -178,3 +180,95 @@ class GroupNorm(Module):
|
||||
)
|
||||
x = group_norm(x)
|
||||
return (self.weight * x + self.bias) if "weight" in self else x
|
||||
|
||||
|
||||
class BatchNorm1d(Module):
|
||||
r"""Applies Batch Normalization [1] to the inputs.
|
||||
|
||||
Computes
|
||||
|
||||
.. math::
|
||||
|
||||
y = \frac{x - E[x]}{\sqrt{Var[x]} + \epsilon} \gamma + \beta,
|
||||
|
||||
where :math:`\gamma` and :math:`\beta` are learned per feature dimension
|
||||
parameters initialized at 1 and 0 respectively.
|
||||
|
||||
[1]: https://arxiv.org/abs/1502.03167
|
||||
|
||||
Args:
|
||||
num_features (int): The feature dimension of the input to normalize over.
|
||||
eps (float, optional): A small additive constant for numerical stability. Default is 1e-5.
|
||||
momentum (float, optional): The momentum for updating the running mean and variance. Default is 0.1.
|
||||
affine (bool, optional): If True, learn an affine transform to apply after the normalization. Default is True.
|
||||
|
||||
Examples:
|
||||
>>> import mlx.core as mx
|
||||
>>> import mlx.nn as nn
|
||||
|
||||
>>> # With Learnable Parameters
|
||||
>>> m = nn.BatchNorm1d(100)
|
||||
>>> # Without Learnable Parameters
|
||||
>>> m = nn.BatchNorm1d(4, affine=False)
|
||||
>>> input = mx.random.normal(20, 4)
|
||||
>>> output = m(input)
|
||||
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
num_features: int,
|
||||
eps: float = 1e-5,
|
||||
momentum: float = 0.1,
|
||||
affine: bool = True,
|
||||
):
|
||||
super().__init__()
|
||||
if affine:
|
||||
self.bias = mx.zeros((num_features,))
|
||||
self.weight = mx.ones((num_features,))
|
||||
|
||||
self.num_features = num_features
|
||||
self.eps = eps
|
||||
self.momentum = momentum
|
||||
self.running_mean = mx.zeros((num_features,))
|
||||
self.running_var = mx.ones((num_features,))
|
||||
|
||||
def _extra_repr(self):
|
||||
return f"num_features={self.num_features}, eps={self.eps}, momentum={self.momentum}, affine={'weight' in self}"
|
||||
|
||||
def _calc_stats(self, x: mx.array) -> Tuple[mx.array, mx.array]:
|
||||
"""
|
||||
Calculate the mean and variance of the input tensor.
|
||||
|
||||
Args:
|
||||
x (mx.array): Input tensor.
|
||||
|
||||
Returns:
|
||||
tuple: Tuple containing mean and variance.
|
||||
"""
|
||||
means = mx.mean(x, axis=0, keepdims=True)
|
||||
var = mx.var(x, axis=0, keepdims=True)
|
||||
self.running_mean = (
|
||||
self.momentum * self.running_mean + (1 - self.momentum) * means
|
||||
)
|
||||
self.running_var = self.momentum * self.running_var + (1 - self.momentum) * var
|
||||
return means, var
|
||||
|
||||
def __call__(self, x: mx.array):
|
||||
"""
|
||||
Forward pass of BatchNorm1d.
|
||||
|
||||
Args:
|
||||
x (mx.array): Input tensor.
|
||||
|
||||
Returns:
|
||||
mx.array: Output tensor.
|
||||
"""
|
||||
if x.ndim != 2:
|
||||
raise ValueError("BatchNorm1d only supports 2D inputs")
|
||||
|
||||
means, var = self.running_mean, self.running_var
|
||||
if self.training:
|
||||
means, var = self._calc_stats(x)
|
||||
x = (x - means) * mx.rsqrt(var + self.eps)
|
||||
return (self.weight * x + self.bias) if "weight" in self else x
|
||||
|
Loading…
Reference in New Issue
Block a user