<p>我目前正在与Scikit Learn合作,在尝试训练高斯HMM时遇到了以下问题:</p>
<blockquote>
<p>File "/Library/Python/2.7/site-packages/sklearn/hmm.py", line 443, in fit </p>
<pre><code>self._do_mstep(stats, self.params)
</code></pre>
<p>File "/Library/Python/2.7/site-packages/sklearn/hmm.py", line 798, in _do_mstep</p>
<pre><code>super(GaussianHMM, self)._do_mstep(stats, params)
</code></pre>
<p>File "/Library/Python/2.7/site-packages/sklearn/hmm.py", line 580, in _do_mstep</p>
<pre><code>np.maximum(self.startprob_prior - 1.0 + stats['start'], 1e-20))
</code></pre>
<p>File "/Library/Python/2.7/site-packages/sklearn/hmm.py", line 476, in _set_startprob</p>
<pre><code>raise ValueError('startprob must sum to 1.0')
</code></pre>
<p>ValueError: startprob must sum to 1.0</p>
</blockquote>
<p>如果我去掉一些特征(每次观察少于13个特征),它仍然有效。我已经检查了所有的输入是否有效,并且只包含数字浮动64S对于每个培训示例。有什么问题吗?
谢谢!在</p>