开发入门
因开发者可能会在OrangePi AIpro（下称：香橙派开发板）进行自定义模型和案例开发，本章节通过基于MindSpore的手写数字识别案例，说明香橙派开发板中的开发注意事项。
环境准备
开发者拿到香橙派开发板后，首先需要进行硬件资源确认、镜像烧录以及CANN和MindSpore版本的升级，才可运行该案例，具体如下：
硬件：香橙派AIpro 16G 8-12T开发板
镜像：香橙派官网Ubuntu镜像
CANN：8.0.RC3.alpha002
MindSpore：2.4.10
镜像烧录
运行该案例需要烧录香橙派官网Ubuntu镜像，参考镜像烧录章节。
CANN升级
参考CANN升级章节。
MindSpore升级
参考MindSpore升级章节。
[1]:
from mindspore import nn
from mindspore.dataset import vision, transforms
from mindspore.dataset import MnistDataset
/home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
setattr(self, word, getattr(machar, word).flat[0])
/home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
return self._float_to_str(self.smallest_subnormal)
/home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
setattr(self, word, getattr(machar, word).flat[0])
/home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
return self._float_to_str(self.smallest_subnormal)
设置运行环境
由于资源限制，需开启性能优化模式，具体设置如下参数：
max_device_memory=“2GB” : 设置设备可用的最大内存为2GB。
mode=mindspore.GRAPH_MODE : 表示在GRAPH_MODE模式中运行。
device_target=“Ascend” : 表示待运行的目标设备为Ascend。
jit_config={“jit_level”:“O2”} : 编译优化级别开启极致性能优化，使用下沉的执行方式。
ascend_config={“precision_mode”:“allow_mix_precision”} : 自动混合精度，自动将部分算子的精度降低到float16或bfloat16。
[2]:
import mindspore
mindspore.set_context(max_device_memory="2GB", mode=mindspore.GRAPH_MODE, device_target="Ascend", jit_config={"jit_level":"O2"}, ascend_config={"precision_mode":"allow_mix_precision"})
数据集准备与加载
MindSpore提供基于Pipeline的数据引擎，通过数据集（Dataset）实现高效的数据预处理。在本案例中，我们使用Mnist数据集，自动下载完成后，使用mindspore.dataset提供的数据变换进行预处理。
[3]:
#install download
!pip install download
Looking in indexes: https://repo.huaweicloud.com/repository/pypi/simple/
Requirement already satisfied: download in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (0.3.5)
Requirement already satisfied: tqdm in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from download) (4.66.5)
Requirement already satisfied: six in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from download) (1.16.0)
Requirement already satisfied: requests in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from download) (2.32.3)
Requirement already satisfied: charset-normalizer<4,>=2 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (3.4.0)
Requirement already satisfied: idna<4,>=2.5 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (3.10)
Requirement already satisfied: urllib3<3,>=1.21.1 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (2.2.3)
Requirement already satisfied: certifi>=2017.4.17 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (2024.8.30)
[4]:
# Download data from open datasets
from download import download
url = "https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/" \
"notebook/datasets/MNIST_Data.zip"
path = download(url, "./", kind="zip", replace=True)
Downloading data from https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/notebook/datasets/MNIST_Data.zip (10.3 MB)
file_sizes: 100%|███████████████████████████| 10.8M/10.8M [00:00<00:00, 101MB/s]
Extracting zip file...
Successfully downloaded / unzipped to ./
MNIST数据集目录结构如下：
MNIST_Data
└── train
├── train-images-idx3-ubyte (60000个训练图片)
├── train-labels-idx1-ubyte (60000个训练标签)
└── test
├── t10k-images-idx3-ubyte (10000个测试图片)
├── t10k-labels-idx1-ubyte (10000个测试标签)
数据下载完成后，获得数据集对象。
[5]:
train_dataset = MnistDataset('MNIST_Data/train')
test_dataset = MnistDataset('MNIST_Data/test')
打印数据集中包含的数据列名，用于dataset的预处理。
[6]:
print(train_dataset.get_col_names())
['image', 'label']
MindSpore的dataset使用数据处理流水线（Data Processing Pipeline），需指定map、batch、shuffle等操作。这里我们使用map对图像数据及标签进行变换处理，将输入的图像缩放为1/255，根据均值0.1307和标准差值0.3081进行归一化处理，然后将处理好的数据集打包为大小为64的batch。
[7]:
def datapipe(dataset, batch_size):
image_transforms = [
vision.Rescale(1.0 / 255.0, 0),
vision.Normalize(mean=(0.1307,), std=(0.3081,)),
vision.HWC2CHW()
]
label_transform = transforms.TypeCast(mindspore.int32)
dataset = dataset.map(image_transforms, 'image')
dataset = dataset.map(label_transform, 'label')
dataset = dataset.batch(batch_size)
return dataset
[8]:
# Map vision transforms and batch dataset
train_dataset = datapipe(train_dataset, 64)
test_dataset = datapipe(test_dataset, 64)
可使用create_tuple_iterator 或create_dict_iterator对数据集进行迭代访问，查看数据和标签的shape和datatype。
[9]:
for image, label in test_dataset.create_tuple_iterator():
print(f"Shape of image [N, C, H, W]: {image.shape} {image.dtype}")
print(f"Shape of label: {label.shape} {label.dtype}")
break
Shape of image [N, C, H, W]: (64, 1, 28, 28) Float32
Shape of label: (64,) Int32
[10]:
for data in test_dataset.create_dict_iterator():
print(f"Shape of image [N, C, H, W]: {data['image'].shape} {data['image'].dtype}")
print(f"Shape of label: {data['label'].shape} {data['label'].dtype}")
break
Shape of image [N, C, H, W]: (64, 1, 28, 28) Float32
Shape of label: (64,) Int32
模型构建
[11]:
# Define model
class Network(nn.Cell):
def __init__(self):
super().__init__()
self.flatten = nn.Flatten()
self.dense_relu_sequential = nn.SequentialCell(
nn.Dense(28*28, 512),
nn.ReLU(),
nn.Dense(512, 512),
nn.ReLU(),
nn.Dense(512, 10)
)
def construct(self, x):
x = self.flatten(x)
logits = self.dense_relu_sequential(x)
return logits
model = Network()
print(model)
Network<
(flatten): Flatten<>
(dense_relu_sequential): SequentialCell<
(0): Dense<input_channels=784, output_channels=512, has_bias=True>
(1): ReLU<>
(2): Dense<input_channels=512, output_channels=512, has_bias=True>
(3): ReLU<>
(4): Dense<input_channels=512, output_channels=10, has_bias=True>
>
>
模型训练
在模型训练中，一个完整的训练过程（step）需要实现以下三步：
正向计算：模型预测结果（logits），并与正确标签（label）求预测损失（loss）。
反向传播：利用自动微分机制，自动求模型参数（parameters）对于loss的梯度（gradients）。
参数优化：将梯度更新到参数上。
MindSpore使用函数式自动微分机制，因此针对上述步骤需要实现：
定义正向计算函数。
使用value_and_grad通过函数变换获得梯度计算函数。
定义训练函数，使用set_train设置为训练模式，执行正向计算、反向传播和参数优化。
[12]:
# Instantiate loss function and optimizer
loss_fn = nn.CrossEntropyLoss()
optimizer = nn.SGD(model.trainable_params(), 1e-2)
# 1. Define forward function
def forward_fn(data, label):
logits = model(data)
loss = loss_fn(logits, label)
return loss, logits
# 2. Get gradient function
grad_fn = mindspore.value_and_grad(forward_fn, None, optimizer.parameters, has_aux=True)
# 3. Define function of one-step training
def train_step(data, label):
(loss, _), grads = grad_fn(data, label)
optimizer(grads)
return loss
def train(model, dataset):
size = dataset.get_dataset_size()
model.set_train()
for batch, (data, label) in enumerate(dataset.create_tuple_iterator()):
loss = train_step(data, label)
if batch % 100 == 0:
loss, current = loss.asnumpy(), batch
print(f"loss: {loss:>7f} [{current:>3d}/{size:>3d}]")
除训练外，我们定义测试函数，用来评估模型的性能。
[13]:
def test(model, dataset, loss_fn):
num_batches = dataset.get_dataset_size()
model.set_train(False)
total, test_loss, correct = 0, 0, 0
for data, label in dataset.create_tuple_iterator():
pred = model(data)
total += len(data)
test_loss += loss_fn(pred, label).asnumpy()
correct += (pred.argmax(1) == label).asnumpy().sum()
test_loss /= num_batches
correct /= total
print(f"Test: \n Accuracy: {(100*correct):>0.1f}%, Avg loss: {test_loss:>8f} \n")
训练过程需多次迭代数据集，一次完整的迭代称为一轮（epoch）。在每一轮，遍历训练集进行训练，结束后使用测试集进行预测。打印每一轮的loss值和预测准确率（Accuracy），可以看到loss在不断下降，Accuracy在不断提高。
[14]:
epochs = 3
for t in range(epochs):
print(f"Epoch {t+1}\n-------------------------------")
train(model, train_dataset)
test(model, test_dataset, loss_fn)
print("Done!")
Epoch 1
-------------------------------
loss: 2.307922 [ 0/938]
loss: 1.746887 [100/938]
loss: 0.848251 [200/938]
loss: 0.607513 [300/938]
loss: 0.369690 [400/938]
loss: 0.382843 [500/938]
loss: 0.293686 [600/938]
loss: 0.391556 [700/938]
loss: 0.227386 [800/938]
loss: 0.189972 [900/938]
Test:
Accuracy: 91.1%, Avg loss: 0.314895
Epoch 2
-------------------------------
loss: 0.346725 [ 0/938]
loss: 0.268921 [100/938]
loss: 0.247742 [200/938]
loss: 0.196686 [300/938]
loss: 0.264954 [400/938]
loss: 0.320938 [500/938]
loss: 0.368820 [600/938]
loss: 0.274811 [700/938]
loss: 0.373581 [800/938]
loss: 0.441010 [900/938]
Test:
Accuracy: 92.8%, Avg loss: 0.247373
Epoch 3
-------------------------------
loss: 0.168976 [ 0/938]
loss: 0.313812 [100/938]
loss: 0.195068 [200/938]
loss: 0.329803 [300/938]
loss: 0.464447 [400/938]
loss: 0.170197 [500/938]
loss: 0.280670 [600/938]
loss: 0.324707 [700/938]
loss: 0.134583 [800/938]
loss: 0.191467 [900/938]
Test:
Accuracy: 93.9%, Avg loss: 0.207696
Done!
保存模型
模型训练完成后，需要将其参数进行保存。
[15]:
# Save checkpoint
mindspore.save_checkpoint(model, "model.ckpt")
print("Saved Model to model.ckpt")
Saved Model to model.ckpt
权重加载
加载保存的权重分为两步：
重新实例化模型对象，构造模型。
加载模型参数，并将其加载至模型上。
[16]:
# Instantiate a random initialized model
model = Network()
# Load checkpoint and load parameter to model
param_dict = mindspore.load_checkpoint("model.ckpt")
param_not_load, _ = mindspore.load_param_into_net(model, param_dict)
print(param_not_load)
[]
param_not_load是未被加载的参数列表，为空时代表所有参数均加载成功。
模型推理
加载后的模型可以直接用于预测推理。
[17]:
import matplotlib.pyplot as plt
model.set_train(False)
for data, label in test_dataset:
pred = model(data)
predicted = pred.argmax(1)
print(f'Predicted: "{predicted[:6]}", Actual: "{label[:6]}"')
# 显示数字及数字的预测值
plt.figure()
for i in range(6):
plt.subplot(2, 3, i + 1)
# 若预测正确，显示为蓝色；若预测错误，显示为红色
color = 'blue' if predicted[i] == label[i] else 'red'
plt.title('Predicted:{}'.format(predicted[i]), color=color)
plt.imshow(data.asnumpy()[i][0], interpolation="None", cmap="gray")
plt.axis('off')
plt.show()
break
Predicted: "[2 1 0 4 1 7]", Actual: "[2 1 0 4 1 7]"
本案例已同步上线GitHub仓，更多案例可参考该仓库。
本案例运行所需环境：
硬件：香橙派AIpro 16G 8-12T开发板
镜像：香橙派官网Ubuntu镜像
CANN：8.0.RC3.alpha002
MindSpore：2.4.10