我正在尝试使用sklearn将GaussianMixture与一堆猫狗图片相匹配。我提供了一个大小不一的数组(50,30000),其中50个数据点(25只猫和25只狗图片),30000是我将每张图片转换为numpy数组并调整为(100,100,3)后的功能数量。它抛出了内存错误。在运行此代码之前,我有4GB的RAM和70%的使用率。任何人都可以建议我如何调试sklearn中GaussianMixture fit方法使用的内存量。或者任何人都可以提供一些代码来批量生产。
以下是代码
print(img_coll_cat_dog.shape)
print(img_coll_cat_dog.nbytes)
print(img_coll_cat_dog.itemsize)
结果:
(50, 30000)
12000000 bytes
8
gmix = mixture.GaussianMixture(n_components=2, covariance_type='full')
gmix.fit(img_coll_cat_dog)
以下是我得到的错误。
MemoryError Traceback (most recent call last)
<ipython-input-32-c0370476a619> in <module>()
1 gmix = mixture.GaussianMixture(n_components=2, covariance_type='full')
----> 2 gmix.fit(img_coll_cat_dog)
~/dl/dl3/lib/python3.5/site-packages/sklearn/mixture/base.py in fit(self, X, y)
205
206 if do_init:
--> 207 self._initialize_parameters(X, random_state)
208 self.lower_bound_ = -np.infty
209
~/dl/dl3/lib/python3.5/site-packages/sklearn/mixture/base.py in _initialize_parameters(self, X, random_state)
155 % self.init_params)
156
--> 157 self._initialize(X, resp)
158
159 @abstractmethod
~/dl/dl3/lib/python3.5/site-packages/sklearn/mixture/gaussian_mixture.py in _initialize(self, X, resp)
629
630 weights, means, covariances = _estimate_gaussian_parameters(
--> 631 X, resp, self.reg_covar, self.covariance_type)
632 weights /= n_samples
633
~/dl/dl3/lib/python3.5/site-packages/sklearn/mixture/gaussian_mixture.py in _estimate_gaussian_parameters(X, resp, reg_covar, covariance_type)
283 "diag": _estimate_gaussian_covariances_diag,
284 "spherical": _estimate_gaussian_covariances_spherical
--> 285 }[covariance_type](resp, X, nk, means, reg_covar)
286 return nk, means, covariances
287
~/dl/dl3/lib/python3.5/site-packages/sklearn/mixture/gaussian_mixture.py in _estimate_gaussian_covariances_full(resp, X, nk, means, reg_covar)
162 """
163 n_components, n_features = means.shape
--> 164 covariances = np.empty((n_components, n_features, n_features))
165 for k in range(n_components):
166 diff = X - means[k]
MemoryError:
非常感谢任何帮助。
答案 0 :(得分:2)
尝试设置covariance_type ='diag'