<p>下载培训/测试图像和标签:</p>
<ul>
<li>train-images-idx3-ubyte.gz:训练集图像</li>
<li>train-labels-idx1-ubyte.gz:训练集标签</li>
<li>t10k-images-idx3-ubyte.gz:测试集图像</li>
<li>t10k-labels-idx1-ubyte.gz:测试集标签</li>
</ul>
<p>在一个工作场所解压,比如<code>samples/</code>。</p>
<p>从PyPi获取<a href="https://github.com/sorki/python-mnist" rel="noreferrer">python-mnist</a>包:</p>
<pre><code>pip install python-mnist
</code></pre>
<p>导入<code>mnist</code>包并读取训练/测试图像:</p>
<pre><code>from mnist import MNIST
mndata = MNIST('samples')
images, labels = mndata.load_training()
# or
images, labels = mndata.load_testing()
</code></pre>
<p>要在控制台上显示图像,请执行以下操作:</p>
<pre><code>index = random.randrange(0, len(images)) # choose an index ;-)
print(mndata.display(images[index]))
</code></pre>
<p>你会得到这样的东西:</p>
<pre><code>............................
............................
............................
............................
............................
.................@@.........
..............@@@@@.........
............@@@@............
..........@@................
..........@.................
...........@................
...........@................
...........@...@............
...........@@@@@.@..........
...........@@@...@@.........
...........@@.....@.........
..................@.........
..................@@........
..................@@........
..................@.........
.................@@.........
...........@.....@..........
...........@....@@..........
............@@@@............
.............@..............
............................
............................
............................
</code></pre>
<p>说明:</p>
<ul>
<li>每个<em>图像</em>列表的<em>图像</em>都是一个无符号字节的Python<code>list</code>。</li>
<li><em>标签</em>是无符号字节的Python<code>array</code>。</li>
</ul>