Python内存分配 - 它可以被释放吗？

Question

我提出prior post关于一个问题，其中我在释放一些变量在python中使用的内存时遇到了一些问题。尝试了很多东西，最后我尝试在代码上使用内存分析器来查看内存的分配位置。

这是内存分析器的输出：

Line #    Mem usage    Increment   Line Contents
================================================
    30   28.523 MiB    0.000 MiB   @profile     
    31                             def store_file(file_name,data):
    32   29.027 MiB    0.504 MiB    with h5py.File(file_name,'w',driver='core',backing_store=False) as f:
    33   28.520 MiB   -0.508 MiB        dset = f.create_dataset("mydataset", (100,) ,dtype='i')


Filename: array.py

Line #    Mem usage    Increment   Line Contents
================================================
    35   24.656 MiB    0.000 MiB   @profile     
    36                             def load_data_overlap(saved):
    37                                 #os.chdir(numpy_train)
    38   24.656 MiB    0.000 MiB       print "Inside function!..."
    39   24.656 MiB    0.000 MiB       if saved == False:
    40   25.531 MiB    0.875 MiB           train_files = np.random.randint(255,size=(1,40,690,4))
    41   25.531 MiB    0.000 MiB           train_input_data_interweawed_normalized = []
    42                                     #print "Storing train pic to numpy"
    43   25.531 MiB    0.000 MiB           part = 0
    44   28.523 MiB    2.992 MiB           for i in xrange(10):
    45   28.523 MiB    0.000 MiB            print resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
    46   28.523 MiB    0.000 MiB            if resource.getrusage(resource.RUSAGE_SELF).ru_maxrss > 27033600:
    47   28.523 MiB    0.000 MiB                print "Max ram storing part: " + str(part) + " At entry: " + str(i)
    48                                          #print resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
    49                                          #print "Storing Train input"
    50   28.523 MiB    0.000 MiB                file_name = 'train_input_'+'part_'+str(part)+'_'+str(dim)+'_'+str(total_frames_with_deltas)+'_window_height_'+str(window_height)+'.h5'
    51   27.492 MiB   -1.031 MiB                store_file(file_name,train_input_data_interweawed_normalized)
    52                                          #print getsizeof(train_input_data_interweawed_normalized)
    53                                          #print resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
    54   27.492 MiB    0.000 MiB                part = part + 1             
    55   27.492 MiB    0.000 MiB                del train_input_data_interweawed_normalized
    56   27.492 MiB    0.000 MiB                gc.collect()
    57                                          #print resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
    58   27.492 MiB    0.000 MiB                train_input_data_interweawed_normalized = []
    59                                          #print getsizeof(train_input_data_interweawed_normalized)
    60                                          #print resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
    61                                          #raw_input("something")
    62   28.523 MiB    1.031 MiB            for plot in train_files:
    63   28.523 MiB    0.000 MiB                overlaps_reshaped = np.random.randint(10,size=(45,200,5,3))
    64   28.523 MiB    0.000 MiB                for ind_plot in overlaps_reshaped.reshape(overlaps_reshaped.shape[1],overlaps_reshaped.shape[0],overlaps_reshaped.shape[2],overlaps_reshaped.shape[3]): 
    65   28.523 MiB    0.000 MiB                    ind_plot_reshaped = ind_plot.reshape(ind_plot.shape[0],1,ind_plot.shape[1],ind_plot.shape[2])
    66   28.523 MiB    0.000 MiB                    train_input_data_interweawed_normalized.append(ind_plot_reshaped)
    67   28.523 MiB    0.000 MiB       print len(train_input_data_interweawed_normalized)

内存被分配到变量中，这些变量总是被清除，但ram的使用从未被释放..

然后，我尝试将内存使用情况随着时间的推移进行可视化，并看到了这一点：

似乎内存使用情况似乎没有返回..看起来python代码动态分配了必要的内存并且从未返回过它。 好像“代码思想” - 它需要那么多内存，因为变量之前使用了这么多内存。所以像矢量这样的东西会增加尺寸，但永远不会减少......

这是正确的理解 - 是python如何管理内存？ 我试过谷歌搜索主题，但大多数帖子建议使用gc.collect() - 从而暗示它应该是可能的吗？..