{"id":940228,"date":"2024-12-26T20:55:57","date_gmt":"2024-12-26T12:55:57","guid":{"rendered":"https:\/\/docs.pingcode.com\/ask\/ask-ask\/940228.html"},"modified":"2024-12-26T20:56:00","modified_gmt":"2024-12-26T12:56:00","slug":"mxnet%e5%a6%82%e4%bd%95%e6%89%a7%e8%a1%8cpython","status":"publish","type":"post","link":"https:\/\/docs.pingcode.com\/ask\/ask-ask\/940228.html","title":{"rendered":"mxnet\u5982\u4f55\u6267\u884cpython"},"content":{"rendered":"

\"mxnet\u5982\u4f55\u6267\u884cpython\"<\/p>\n

\u8981\u5728MXNet\u4e2d\u6267\u884cPython\u4ee3\u7801\uff0c\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u51e0\u4e2a\u6b65\u9aa4\uff1a\u5b89\u88c5MXNet\u5e93\u3001\u5bfc\u5165MXNet\u6a21\u5757\u3001\u5b9a\u4e49\u548c\u521d\u59cb\u5316\u7b26\u53f7\u3001\u521b\u5efa\u6267\u884c\u5668\u5e76\u8fd0\u884c\u3001\u4f7f\u7528NDArray\u8fdb\u884c\u6570\u636e\u64cd\u4f5c\u3002\u5b89\u88c5MXNet\u5e93\u662f\u6700\u57fa\u7840\u7684\u4e00\u6b65\uff0c\u901a\u5e38\u901a\u8fc7pip\u547d\u4ee4\u5373\u53ef\u5b8c\u6210\u3002<\/strong><\/p>\n<\/p>\n

\u4e00\u3001\u5b89\u88c5MXNet\u5e93<\/p>\n<\/p>\n

\u5728\u5f00\u59cb\u4f7f\u7528MXNet\u4e4b\u524d\uff0c\u9996\u5148\u9700\u8981\u5b89\u88c5\u8be5\u5e93\u3002MXNet\u7684\u5b89\u88c5\u53ef\u4ee5\u901a\u8fc7pip\u547d\u4ee4\u6765\u5b8c\u6210\uff0c\u4e5f\u53ef\u4ee5\u6839\u636e\u5177\u4f53\u9700\u6c42\u9009\u62e9CPU\u7248\u672c\u6216GPU\u7248\u672c\u3002\u4e00\u822c\u6765\u8bf4\uff0c\u5982\u679c\u4f60\u4ec5\u4ec5\u662f\u4e3a\u4e86\u5b66\u4e60\u76ee\u7684\u5e76\u4e14\u4e0d\u9700\u8981\u8fdb\u884c\u5927\u89c4\u6a21\u8ba1\u7b97\uff0c\u53ef\u4ee5\u9009\u62e9\u5b89\u88c5CPU\u7248\u672c\u3002\u5b89\u88c5\u547d\u4ee4\u5982\u4e0b\uff1a<\/p>\n<\/p>\n

pip install mxnet<\/p>\n
\u6216\u8005\u5b89\u88c5GPU\u7248\u672c<\/strong><\/h2>\n
pip install mxnet-cu101<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u5b89\u88c5\u5b8c\u6210\u540e\uff0c\u53ef\u4ee5\u901a\u8fc7\u5bfc\u5165mxnet\u6a21\u5757\u6765\u9a8c\u8bc1\u5b89\u88c5\u662f\u5426\u6210\u529f\uff1a<\/p>\n<\/p>\n
import mxnet as mx<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e8c\u3001\u5bfc\u5165MXNet\u6a21\u5757<\/p>\n<\/p>\n
\u5728\u7f16\u5199Python\u4ee3\u7801\u65f6\uff0c\u5bfc\u5165MXNet\u6a21\u5757\u662f\u5fc5\u987b\u7684\u3002MXNet\u63d0\u4f9b\u4e86\u591a\u79cd\u6a21\u5757\u7528\u4e8e\u4e0d\u540c\u7684\u4efb\u52a1\uff0c\u6700\u5e38\u7528\u7684\u5305\u62ecmx.nd<\/code>\u7528\u4e8e\u5904\u7406\u591a\u7ef4\u6570\u7ec4\uff0cmx.sym<\/code>\u7528\u4e8e\u5b9a\u4e49\u7b26\u53f7\u5316\u7684\u8ba1\u7b97\u56fe\uff0cmx.gluon<\/code>\u7528\u4e8e\u9ad8\u7ea7API\u6784\u5efa\u795e\u7ecf\u7f51\u7edc\u3002<\/p>\n<\/p>\n
from mxnet import nd, sym, gluon<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e09\u3001\u5b9a\u4e49\u548c\u521d\u59cb\u5316\u7b26\u53f7<\/p>\n<\/p>\n
MXNet\u7684\u7b26\u53f7\uff08Symbol\uff09\u8868\u793a\u8ba1\u7b97\u56fe\u4e2d\u7684\u4e00\u4e2a\u8282\u70b9\uff0c\u7528\u4e8e\u63cf\u8ff0\u8ba1\u7b97\u8fc7\u7a0b\u3002\u53ef\u4ee5\u4f7f\u7528mx.sym<\/code>\u6a21\u5757\u6765\u5b9a\u4e49\u7b26\u53f7\u5316\u7684\u8ba1\u7b97\u56fe\u3002\u4ee5\u4e0b\u662f\u4e00\u4e2a\u7b80\u5355\u7684\u7ebf\u6027\u56de\u5f52\u6a21\u578b\u7684\u4f8b\u5b50\uff1a<\/p>\n<\/p>\n
# \u5b9a\u4e49\u7b26\u53f7\u53d8\u91cf<\/p>\n
x = sym.Variable('data')<\/p>\n
w = sym.Variable('weight')<\/p>\n
b = sym.Variable('bias')<\/p>\n
\u5b9a\u4e49\u7ebf\u6027\u6a21\u578b<\/strong><\/h2>\n
linear_model = sym.broadcast_add(sym.dot(x, w), b)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u5b9a\u4e49\u597d\u7b26\u53f7\u540e\uff0c\u901a\u5e38\u9700\u8981\u521d\u59cb\u5316\u53c2\u6570\u3002\u5728MXNet\u4e2d\uff0c\u53ef\u4ee5\u901a\u8fc7mx.init<\/code>\u6a21\u5757\u8fdb\u884c\u521d\u59cb\u5316\uff1a<\/p>\n<\/p>\n
import mxnet.ndarray as nd<\/p>\n
from mxnet import init<\/p>\n
w = nd.random.normal(shape=(2, 1))<\/p>\n
b = nd.random.normal(shape=(1,))<\/p>\n
init.Normal()(w)<\/p>\n
init.Normal()(b)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u56db\u3001\u521b\u5efa\u6267\u884c\u5668\u5e76\u8fd0\u884c<\/p>\n<\/p>\n
\u5728\u5b9a\u4e49\u597d\u8ba1\u7b97\u56fe\u540e\uff0c\u9700\u8981\u521b\u5efa\u4e00\u4e2a\u6267\u884c\u5668\uff08Executor\uff09\u6765\u6267\u884c\u8ba1\u7b97\u3002\u901a\u8fc7\u7b26\u53f7\u7684bind<\/code>\u65b9\u6cd5\u53ef\u4ee5\u5c06\u7b26\u53f7\u7ed1\u5b9a\u5230\u5177\u4f53\u7684\u6570\u636e\u4e0a\uff1a<\/p>\n<\/p>\n
executor = linear_model.bind(ctx=mx.cpu(), args={'data': x_data, 'weight': w, 'bias': b})<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u7136\u540e\u53ef\u4ee5\u901a\u8fc7executor.forward()<\/code>\u65b9\u6cd5\u6267\u884c\u8ba1\u7b97\uff1a<\/p>\n<\/p>\n
output = executor.forward()<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e94\u3001\u4f7f\u7528NDArray\u8fdb\u884c\u6570\u636e\u64cd\u4f5c<\/p>\n<\/p>\n
MXNet\u7684NDArray<\/code>\u662f\u7528\u4e8e\u5b58\u50a8\u548c\u64cd\u4f5c\u591a\u7ef4\u6570\u7ec4\u7684\u6570\u636e\u7ed3\u6784\uff0c\u7c7b\u4f3c\u4e8eNumPy\u7684ndarray<\/code>\u3002\u53ef\u4ee5\u4f7f\u7528NDArray<\/code>\u8fdb\u884c\u5404\u79cd\u7ebf\u6027\u4ee3\u6570\u64cd\u4f5c\u3001\u968f\u673a\u6570\u751f\u6210\u548c\u5176\u4ed6\u79d1\u5b66\u8ba1\u7b97\u3002<\/p>\n<\/p>\n
import mxnet.ndarray as nd<\/p>\n
\u521b\u5efa\u4e00\u4e2a\u968f\u673a\u6570\u7ec4<\/strong><\/h2>\n
a = nd.random.normal(shape=(3, 3))<\/p>\n
\u77e9\u9635\u52a0\u6cd5<\/strong><\/h2>\n
b = nd.ones((3, 3))<\/p>\n
c = a + b<\/p>\n
\u77e9\u9635\u4e58\u6cd5<\/strong><\/h2>\n
d = nd.dot(a, b)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u516d\u3001\u4f7f\u7528Gluon\u6784\u5efa\u795e\u7ecf\u7f51\u7edc<\/p>\n<\/p>\n
MXNet\u7684Gluon\u6a21\u5757\u63d0\u4f9b\u4e86\u4e00\u79cd\u66f4\u9ad8\u7ea7\u7684API\uff0c\u7528\u4e8e\u5feb\u901f\u6784\u5efa\u548c\u8bad\u7ec3\u795e\u7ecf\u7f51\u7edc\u3002\u53ef\u4ee5\u901a\u8fc7\u7ee7\u627fgluon.nn.Block<\/code>\u7c7b\u6765\u5b9a\u4e49\u7f51\u7edc\u7ed3\u6784\uff1a<\/p>\n<\/p>\n
from mxnet.gluon import nn<\/p>\n
class MyNet(nn.Block):<\/p>\n
    def __init__(self):<\/p>\n
        super(MyNet, self).__init__()<\/p>\n
        self.dense0 = nn.Dense(128, activation='relu')<\/p>\n
        self.dense1 = nn.Dense(64, activation='relu')<\/p>\n
        self.dense2 = nn.Dense(10)<\/p>\n
    def forward(self, x):<\/p>\n
        x = self.dense0(x)<\/p>\n
        x = self.dense1(x)<\/p>\n
        x = self.dense2(x)<\/p>\n
        return x<\/p>\n
net = MyNet()<\/p>\n
net.initialize()<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e03\u3001\u8bad\u7ec3\u6a21\u578b<\/p>\n<\/p>\n
\u5728\u5b9a\u4e49\u597d\u7f51\u7edc\u7ed3\u6784\u540e\uff0c\u9700\u8981\u8fdb\u884c\u6a21\u578b\u8bad\u7ec3\u3002MXNet\u63d0\u4f9b\u4e86TrAI<\/a>ner<\/code>\u7c7b\u7528\u4e8e\u4f18\u5316\u6a21\u578b\u53c2\u6570\u3002\u53ef\u4ee5\u901a\u8fc7trainer.step(batch_size)<\/code>\u6765\u66f4\u65b0\u53c2\u6570\uff1a<\/p>\n<\/p>\n
from mxnet import autograd, gluon<\/p>\n
\u5b9a\u4e49\u635f\u5931\u51fd\u6570<\/strong><\/h2>\n
loss_fn = gluon.loss.SoftmaxCrossEntropyLoss()<\/p>\n
\u5b9a\u4e49\u4f18\u5316\u5668<\/strong><\/h2>\n
trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': 0.01})<\/p>\n
\u6a21\u62df\u8bad\u7ec3\u6570\u636e<\/strong><\/h2>\n
x_data = nd.random.normal(shape=(10, 2))<\/p>\n
y_data = nd.random.randint(low=0, high=10, shape=(10,))<\/p>\n
\u8bad\u7ec3\u5faa\u73af<\/strong><\/h2>\n
for epoch in range(10):<\/p>\n
    with autograd.record():<\/p>\n
        output = net(x_data)<\/p>\n
        loss = loss_fn(output, y_data)<\/p>\n
    loss.backward()<\/p>\n
    trainer.step(batch_size=10)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u516b\u3001\u8bc4\u4f30\u6a21\u578b<\/p>\n<\/p>\n
\u5728\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u9700\u8981\u8bc4\u4f30\u6a21\u578b\u7684\u8868\u73b0\u3002\u53ef\u4ee5\u901a\u8fc7\u8ba1\u7b97\u6a21\u578b\u5728\u6d4b\u8bd5\u6570\u636e\u4e0a\u7684\u51c6\u786e\u6027\u6765\u8bc4\u4f30\uff1a<\/p>\n<\/p>\n
from mxnet.metric import Accuracy<\/p>\n
metric = Accuracy()<\/p>\n
metric.update(labels=y_data, preds=output)<\/p>\n
print('Accuracy:', metric.get()[1])<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e5d\u3001\u4f7f\u7528GPU\u8fdb\u884c\u52a0\u901f<\/p>\n<\/p>\n
\u5982\u679c\u4f60\u7684\u8ba1\u7b97\u673a\u914d\u5907\u4e86\u652f\u6301CUDA\u7684GPU\uff0c\u53ef\u4ee5\u901a\u8fc7\u6307\u5b9a\u4e0a\u4e0b\u6587\uff08ctx<\/code>\uff09\u6765\u5229\u7528GPU\u52a0\u901f\u8ba1\u7b97\uff1a<\/p>\n<\/p>\n
ctx = mx.gpu() if mx.context.num_gpus() > 0 else mx.cpu()<\/p>\n
\u5c06\u6570\u636e\u548c\u6a21\u578b\u53c2\u6570\u79fb\u52a8\u5230GPU\u4e0a<\/strong><\/h2>\n
x_data = x_data.as_in_context(ctx)<\/p>\n
y_data = y_data.as_in_context(ctx)<\/p>\n
net.initialize(ctx=ctx)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u5341\u3001\u4fdd\u5b58\u548c\u52a0\u8f7d\u6a21\u578b<\/p>\n<\/p>\n
\u5728\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u53ef\u4ee5\u5c06\u6a21\u578b\u53c2\u6570\u4fdd\u5b58\u5230\u6587\u4ef6\u4e2d\uff0c\u4ee5\u4fbf\u540e\u7eed\u4f7f\u7528\u6216\u90e8\u7f72\uff1a<\/p>\n<\/p>\n
net.save_parameters('mynet.params')<\/p>\n
\u52a0\u8f7d\u6a21\u578b\u53c2\u6570<\/strong><\/h2>\n
net.load_parameters('mynet.params', ctx=ctx)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u4f60\u53ef\u4ee5\u5728MXNet\u4e2d\u6267\u884cPython\u4ee3\u7801\u8fdb\u884c\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u7684\u5f00\u53d1\u548c\u8bad\u7ec3\u3002MXNet\u7684\u7075\u6d3b\u6027\u548c\u9ad8\u6548\u6027\u4f7f\u5176\u6210\u4e3a\u6784\u5efa\u548c\u90e8\u7f72\u5927\u89c4\u6a21\u6df1\u5ea6\u5b66\u4e60\u5e94\u7528\u7684\u7406\u60f3\u9009\u62e9\u3002\u5e0c\u671b\u8fd9\u7bc7\u6587\u7ae0\u80fd\u5e2e\u52a9\u4f60\u66f4\u597d\u5730\u7406\u89e3\u5982\u4f55\u5728MXNet\u4e2d\u6267\u884cPython\u4ee3\u7801\u3002<\/p>\n<\/p>\n
\u76f8\u5173\u95ee\u7b54FAQs\uff1a<\/strong><\/h2>\n
 MXNet\u652f\u6301\u54ea\u4e9b\u7248\u672c\u7684Python\uff1f<\/strong>
MXNet\u652f\u6301\u591a\u4e2a\u7248\u672c\u7684Python\uff0c\u5305\u62ecPython 2.7\u548cPython 3.5\u53ca\u4ee5\u4e0a\u7248\u672c\u3002\u5efa\u8bae\u4f7f\u7528\u6700\u65b0\u7684Python 3.x\u7248\u672c\uff0c\u4ee5\u786e\u4fdd\u6700\u4f73\u7684\u6027\u80fd\u548c\u517c\u5bb9\u6027\u3002\u6b64\u5916\uff0c\u786e\u4fdd\u5b89\u88c5\u7684MXNet\u7248\u672c\u4e0e\u6240\u4f7f\u7528\u7684Python\u7248\u672c\u76f8\u5339\u914d\u3002<\/p>\n
\u5728Windows\u548cLinux\u4e0a\u5982\u4f55\u5b89\u88c5MXNet\u4ee5\u4fbf\u4f7f\u7528Python\uff1f<\/strong>
\u5728Windows\u7cfb\u7edf\u4e2d\uff0c\u53ef\u4ee5\u901a\u8fc7\u4f7f\u7528pip\u547d\u4ee4\u5b89\u88c5MXNet\uff0c\u5177\u4f53\u547d\u4ee4\u4e3apip install mxnet<\/code>\u3002\u5bf9\u4e8eLinux\u7528\u6237\uff0c\u5b89\u88c5\u8fc7\u7a0b\u7c7b\u4f3c\uff0c\u53ef\u4ee5\u4f7f\u7528\u76f8\u540c\u7684pip\u547d\u4ee4\uff0c\u6216\u8005\u4f7f\u7528conda\u8fdb\u884c\u73af\u5883\u7ba1\u7406\uff0c\u547d\u4ee4\u4e3aconda install -c conda-forge mxnet<\/code>\u3002\u786e\u4fdd\u5728\u5b89\u88c5\u524d\u66f4\u65b0pip\u5de5\u5177\uff0c\u4ee5\u907f\u514d\u6f5c\u5728\u7684\u4f9d\u8d56\u95ee\u9898\u3002<\/p>\n
\u5982\u4f55\u5728MXNet\u4e2d\u4f7f\u7528Python\u8fdb\u884c\u6a21\u578b\u8bad\u7ec3\uff1f<\/strong>
\u5728MXNet\u4e2d\u8fdb\u884c\u6a21\u578b\u8bad\u7ec3\u9700\u8981\u5148\u5b9a\u4e49\u7f51\u7edc\u67b6\u6784\uff0c\u7136\u540e\u51c6\u5907\u6570\u636e\u5e76\u8bbe\u5b9a\u8bad\u7ec3\u53c2\u6570\u3002\u4f7f\u7528Python\u7684MXNet API\uff0c\u53ef\u4ee5\u901a\u8fc7mxnet.gluon<\/code>\u6a21\u5757\u6784\u5efa\u6a21\u578b\u3002\u7528\u6237\u9700\u8981\u52a0\u8f7d\u6570\u636e\u96c6\uff0c\u4f7f\u7528DataLoader<\/code>\u7c7b\u8fdb\u884c\u6279\u91cf\u5904\u7406\uff0c\u5e76\u4f7f\u7528Trainer<\/code>\u7c7b\u6765\u8bbe\u7f6e\u4f18\u5316\u5668\u548c\u8bad\u7ec3\u6b65\u9aa4\u3002\u53ef\u4ee5\u53c2\u8003MXNet\u7684\u5b98\u65b9\u6587\u6863\u548c\u793a\u4f8b\u4ee3\u7801\u6765\u4e86\u89e3\u8be6\u7ec6\u7684\u5b9e\u73b0\u65b9\u5f0f\u548c\u6700\u4f73\u5b9e\u8df5\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"\u8981\u5728MXNet\u4e2d\u6267\u884cPython\u4ee3\u7801\uff0c\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u51e0\u4e2a\u6b65\u9aa4\uff1a\u5b89\u88c5MXNet\u5e93\u3001\u5bfc\u5165MXNet\u6a21\u5757\u3001\u5b9a\u4e49\u548c\u521d\u59cb\u5316 […]","protected":false},"author":3,"featured_media":940229,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[37],"tags":[],"acf":[],"_links":{"self":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/940228"}],"collection":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/comments?post=940228"}],"version-history":[{"count":"1","href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/940228\/revisions"}],"predecessor-version":[{"id":940230,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/940228\/revisions\/940230"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media\/940229"}],"wp:attachment":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media?parent=940228"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/categories?post=940228"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/tags?post=940228"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}