Matplotlib memory leak -

i have loop turns financial ohlc price data rgb pixel data. process is:

1, ohlc price data

2, draw ohlc data images matplotlib finance

3, rgb pixel data fig.canvas

for program, drawing 50 price data 1 chart, , stacking 4 incremental chart one, getting (120*120*4) per pixel data. since result in opening , closing thousands of matplotlib subplots, having significant memory leak.

i have tried methods deal leak, including:

1, plt.close('all'),

2, del fig, ax

3, using fig = figure.figure(), instead of fig,ax = plt.subplots()

i tried changing matplotlib backend qt4agg agg, leak still exists. tried multiprocessing process, still won't solve problem.

the code here:

    def get_data(file_dir):       data = pd.dataframe.from_csv(file_dir)       return data      file_dir = 'c:/users/czzis/desktop/data/eurusd_all/dat_mt_eurusd_m1_2016.csv'      data = get_data(file_dir)      b = []     start = 0     idx = 50     stack = 4      in range(22370):             = np.zeros(shape = (120,120,1))          in range(stack):              opens = data.iloc[start+i : idx+i ,1].as_matrix()             highs = data.iloc[start+i : idx+i ,2].as_matrix()             lows = data.iloc[start+i : idx+i ,3].as_matrix()             closes = data.iloc[start+i : idx+i ,4].as_matrix()              # draw chart first , rgb data      #       fig = figure.figure()     #       ax = fig.add_axes([1, 1, 1, 1])     #       ax.set_axis_off()              fig, ax = plt.subplots()             plt.axis('off')             matplotlib.finance.candlestick2_ohlc(ax, opens, highs, lows, closes, width=0.6, colorup='w', colordown='k', alpha=1)     #       print(i)     #       del fig     #       canvas = figurecanvas(fig)      #       fig.set_canvas(canvas)              fig.canvas.draw()                observation = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep='')             observation = observation.reshape(fig.canvas.get_width_height()[::-1] + (3,))                     # process data             observation = skimage.transform.resize(observation,(120,120))               observation = np.delete(observation,np.s_[1:4], axis = 2)               observation = np.reshape(observation,(120,120,1))   # final shrinked image data need               = np.append(a,observation,axis = 2)        #       matplotlib.pyplot.close('all')     #       plt.clf()             plt.show()             plt.close('all')     #       del fig,ax     #       plt.close("all")     #       del fig,ax     #       print()     #       del fig.canvas          observation = np.delete(a,np.s_[0:1],axis = 2)         start += 1         idx   += 1          b.append(observation)     

none of method can fix leak. me out? thanks!