File size: 8,159 Bytes
4c27541
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
"""Numpy reference forward pass of a ResNet-20 image classifier (model_spec.json describes it).

Requires numpy only. Call ``forward(params, x_uint8)``:

* ``params``: dict of float arrays keyed by parameter name, PyTorch layout (conv weight [out, in, kh, kw]).
* ``x_uint8``: raw uint8 images of shape [N, *INPUT_SHAPE] (NCHW).
* returns float32 logits [N, N_CLASSES]; the prediction is the argmax over axis 1.

Preprocessing: x / 255, then per channel (x - MEAN[c]) / STD[c].

Architecture (fixed; ``expected_shapes()`` lists every parameter, and any other name or shape is refused):

* stem: ``conv1`` (3x3, stride 1, padding 1, no bias), batch norm ``bn1``, ReLU.
* three stages ``layer1`` .. ``layer3`` of BLOCKS_PER_STAGE basic blocks each, with STAGE_CHANNELS channels. The first
  block of ``layer2`` and of ``layer3`` has stride 2; every other convolution has stride 1. A basic block ``layerL.i``
  computes out = ReLU(bn2(conv2(ReLU(bn1(conv1(x))))) + shortcut(x)), with 3x3 convolutions (padding 1, no bias).
  The shortcut has no parameters. It is the identity, except in ``layer2.0`` and ``layer3.0``, where it keeps every
  second row and column (x[:, :, ::2, ::2]) and pads the channels with zeros, out_channels // 4 on each side.
* global average pooling over the spatial positions gives the features f [N, 64].
* linear head: logits = f @ ``fc.weight``.T + ``fc.bias``, ``fc.weight`` [N_CLASSES, 64] and ``fc.bias``
  [N_CLASSES].

Convolution is 2-D cross-correlation as in torch.nn.functional.conv2d: out[n, o, i, j] = sum_{c, u, v} weight[o, c, u,
v] * xpad[n, c, i * stride + u, j * stride + v], with xpad zero-padded on each spatial side. Batch norm is in
evaluation mode: (x - running_mean) / sqrt(running_var + BN_EPS) * weight + bias, per channel. Convolutions use im2col.
All arithmetic runs in float64 and the logits are cast once to float32. Inputs are processed in chunks of at most
CHUNK images, which bounds memory (about 0.6 GB).
"""

from __future__ import annotations

import numpy as np

INPUT_SHAPE = (3, 32, 32)
MEAN = np.array([0.490903318493413, 0.4817690980200674, 0.4460368943397672], dtype=np.float64)
STD = np.array([0.24706968464047946, 0.24352667019453098, 0.26164691226372544], dtype=np.float64)
N_CLASSES = 10
BN_EPS = 1e-05
STEM_CHANNELS = 16
STAGE_CHANNELS = (16, 32, 64)
STAGE_STRIDES = (1, 2, 2)
BLOCKS_PER_STAGE = 3
BN_FIELDS = ("weight", "bias", "running_mean", "running_var")
CHUNK = 256


def _bn_shapes(name: str, channels: int) -> dict[str, tuple[int, ...]]:
    return {f"{name}.{field}": (channels,) for field in BN_FIELDS}


def _blocks() -> list[tuple[str, int, int, int, bool]]:
    """(name, in channels, out channels, stride, has the subsampling shortcut) of every basic block in forward order."""
    blocks, channels_in = [], STEM_CHANNELS
    for stage, (channels, stride) in enumerate(zip(STAGE_CHANNELS, STAGE_STRIDES), start=1):
        for index in range(BLOCKS_PER_STAGE):
            block_stride = stride if index == 0 else 1
            subsample = block_stride != 1 or channels_in != channels
            blocks.append((f"layer{stage}.{index}", channels_in, channels, block_stride, subsample))
            channels_in = channels
    return blocks


def expected_shapes() -> dict[str, tuple[int, ...]]:
    """Every parameter name of the model with its shape, in forward order."""
    shapes: dict[str, tuple[int, ...]] = {"conv1.weight": (STEM_CHANNELS, INPUT_SHAPE[0], 3, 3)}
    shapes.update(_bn_shapes("bn1", STEM_CHANNELS))
    for name, channels_in, channels, _stride, _subsample in _blocks():
        shapes[f"{name}.conv1.weight"] = (channels, channels_in, 3, 3)
        shapes.update(_bn_shapes(f"{name}.bn1", channels))
        shapes[f"{name}.conv2.weight"] = (channels, channels, 3, 3)
        shapes.update(_bn_shapes(f"{name}.bn2", channels))
    shapes["fc.weight"] = (N_CLASSES, STAGE_CHANNELS[-1])
    shapes["fc.bias"] = (N_CLASSES,)
    return shapes


def _checked(params: dict) -> dict[str, np.ndarray]:
    expected = expected_shapes()
    names = {str(key) for key in params}
    if names != set(expected):
        missing, extra = sorted(set(expected) - names), sorted(names - set(expected))
        raise KeyError(f"parameter names differ from the architecture: missing {missing}, unexpected {extra}")
    tensors = {}
    for name, shape in expected.items():
        value = np.asarray(params[name])
        if not np.issubdtype(value.dtype, np.floating):
            raise TypeError(f"{name} must be a float array, got {value.dtype}")
        if tuple(value.shape) != shape:
            raise ValueError(f"{name} has shape {tuple(value.shape)}, expected {shape}")
        tensors[name] = value.astype(np.float64)
    return tensors


def _conv2d(h: np.ndarray, weight: np.ndarray, stride: int) -> np.ndarray:
    n, channels, height, width = h.shape
    out_channels, in_channels, kh, kw = weight.shape
    if in_channels != channels:
        raise ValueError(f"conv weight {weight.shape} does not fit input {h.shape}")
    pad = kh // 2
    padded = np.pad(h, ((0, 0), (0, 0), (pad, pad), (pad, pad))) if pad else h
    h_out = (height + 2 * pad - kh) // stride + 1
    w_out = (width + 2 * pad - kw) // stride + 1
    windows = np.lib.stride_tricks.sliding_window_view(padded, (kh, kw), axis=(2, 3))
    windows = windows[:, :, : (h_out - 1) * stride + 1 : stride, : (w_out - 1) * stride + 1 : stride]
    cols = windows.transpose(0, 2, 3, 1, 4, 5).reshape(n * h_out * w_out, channels * kh * kw)
    out = cols @ weight.reshape(out_channels, channels * kh * kw).T
    return out.reshape(n, h_out, w_out, out_channels).transpose(0, 3, 1, 2)


def _bn(h: np.ndarray, tensors: dict[str, np.ndarray], name: str) -> np.ndarray:
    scale = tensors[f"{name}.weight"] / np.sqrt(tensors[f"{name}.running_var"] + BN_EPS)
    shift = tensors[f"{name}.bias"] - tensors[f"{name}.running_mean"] * scale
    return h * scale.reshape(1, -1, 1, 1) + shift.reshape(1, -1, 1, 1)


def _subsample(h: np.ndarray, channels: int) -> np.ndarray:
    pad = channels // 4
    return np.pad(h[:, :, ::2, ::2], ((0, 0), (pad, pad), (0, 0), (0, 0)))


def features(params: dict[str, np.ndarray], x_uint8: np.ndarray) -> np.ndarray:
    """float64 features f [N, 64] (the pooled encoder output, before the head) for raw uint8 images."""
    x = _images(x_uint8)
    tensors = _checked(params)
    if x.shape[0] == 0:
        return np.zeros((0, STAGE_CHANNELS[-1]), dtype=np.float64)
    return np.concatenate([_features(tensors, x[start : start + CHUNK]) for start in range(0, x.shape[0], CHUNK)])


def _features(tensors: dict[str, np.ndarray], x: np.ndarray) -> np.ndarray:
    h = x.astype(np.float64) / 255.0
    h = (h - MEAN.reshape(1, -1, 1, 1)) / STD.reshape(1, -1, 1, 1)
    h = np.maximum(_bn(_conv2d(h, tensors["conv1.weight"], 1), tensors, "bn1"), 0.0)
    for name, _channels_in, channels, stride, subsample in _blocks():
        shortcut = _subsample(h, channels) if subsample else h
        out = np.maximum(_bn(_conv2d(h, tensors[f"{name}.conv1.weight"], stride), tensors, f"{name}.bn1"), 0.0)
        out = _bn(_conv2d(out, tensors[f"{name}.conv2.weight"], 1), tensors, f"{name}.bn2")
        h = np.maximum(out + shortcut, 0.0)
    return h.mean(axis=(2, 3))


def _images(x_uint8: np.ndarray) -> np.ndarray:
    x = np.asarray(x_uint8)
    if x.dtype != np.uint8:
        raise TypeError(f"images must be uint8, got {x.dtype}")
    if x.ndim != 4 or tuple(x.shape[1:]) != INPUT_SHAPE:
        raise ValueError(f"images must have shape [N, {', '.join(map(str, INPUT_SHAPE))}], got {list(x.shape)}")
    return x


def forward(params: dict[str, np.ndarray], x_uint8: np.ndarray) -> np.ndarray:
    """Logits float32 [N, N_CLASSES] for raw uint8 images [N, *INPUT_SHAPE]."""
    x = _images(x_uint8)
    tensors = _checked(params)
    if x.shape[0] == 0:
        return np.zeros((0, N_CLASSES), dtype=np.float32)
    chunks = []
    for start in range(0, x.shape[0], CHUNK):
        f = _features(tensors, x[start : start + CHUNK])
        chunks.append((f @ tensors["fc.weight"].T + tensors["fc.bias"]).astype(np.float32))
    return np.concatenate(chunks, axis=0)