grayImg = cv2.imread('../my-images/' + row[1], 0)
keypoints, descriptors = sift.detectAndCompute(grayImg, None)
descriptors = np.insert(descriptors, 0, 128, 1)
vectorArray = np.concatenate((vectorArray, descriptors), axis = 0)

vectorArray.astype(np.int32).tofile('./my_sift_descriptors.fvecs')

#the above code is faulty and should first insert an int32 of with the dimension as value and subsequently float32 values of the vector itself
#the following code is correct:

#assuming you calculated descriptors with SIFT and have opened an output file with 'wb' flag
for descriptor in descriptors:
  dimension_array = array('i', [128])
  dimension_array.tofile(output_file)
  float_array = array('f', descriptor)
  float_array.tofile(output_file)