Apply into NumpyΒΆ
In following parts, we will show some demos about how to use TreeValue
in practice.
For example, now we have a group of structed data in python-dict type, we want to do different operations on differnent key-value pairs inplace, get some statistics such as mean value and task some slices.
In normal cases, we need to unroll multiple for-loop
and if-else
to implement cooresponding operations on each values, and declare additional
temporal variables to save result. All the mentioned contents are executed serially, like the next code examples:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 | import numpy as np T, B = 3, 4 def without_treevalue(batch_): mean_b_list = [] even_index_a_list = [] for i in range(len(batch_)): for k, v in batch_[i].items(): if k == 'a': v = v.astype(np.float32) even_index_a_list.append(v[::2]) elif k == 'b': v = v.astype(np.float32) transformed_v = np.power(v, 2) + 1.0 mean_b_list.append(transformed_v.mean()) elif k == 'c': for k1, v1 in v.items(): if k1 == 'd': v1 = v1.astype(np.float32) else: print('ignore keys: {}'.format(k1)) else: print('ignore keys: {}'.format(k)) for i in range(len(batch_)): for k in batch_[i].keys(): if k == 'd': batch_[i][k]['noise'] = np.random.random(size=(3, 4, 5)) mean_b = sum(mean_b_list) / len(mean_b_list) even_index_a = np.stack(even_index_a_list, axis=0) return batch_, mean_b, even_index_a |
However, with the help of TreeValue
, all the contents mentioned above can be implemented gracefully and efficiently. Users only need to func_treelize
the primitive
numpy functions and pack data with FastTreeValue
, then execute desired operations just like using standard numpy array.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 | import numpy as np from treevalue import FastTreeValue T, B = 3, 4 power = FastTreeValue.func()(np.power) stack = FastTreeValue.func(subside=True)(np.stack) split = FastTreeValue.func(rise=True)(np.split) def with_treevalue(batch_): batch_ = [FastTreeValue(b) for b in batch_] batch_ = stack(batch_) batch_ = batch_.astype(np.float32) batch_.b = power(batch_.b, 2) + 1.0 batch_.c.noise = np.random.random(size=(B, 3, 4, 5)) mean_b = batch_.b.mean() even_index_a = batch_.a[:, ::2] batch_ = split(batch_, indices_or_sections=B, axis=0) return batch_, mean_b, even_index_a |
And we can run these two demos for comparison:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 | import copy import numpy as np from with_treevalue import with_treevalue from without_treevalue import without_treevalue T, B = 3, 4 def get_data(): return { 'a': np.random.random(size=(T, 8)), 'b': np.random.random(size=(6,)), 'c': { 'd': np.random.randint(0, 10, size=(1,)) } } if __name__ == "__main__": batch = [get_data() for _ in range(B)] batch0, mean0, even_index_a0 = without_treevalue(copy.deepcopy(batch)) batch1, mean1, even_index_a1 = with_treevalue(copy.deepcopy(batch)) assert np.abs(mean0 - mean1) < 1e-6 print('mean0 & mean1:', mean0, mean1) print('\n') assert np.abs((even_index_a0 - even_index_a1).max()) < 1e-6 print('even_index_a0:', even_index_a0) print('even_index_a1:', even_index_a1) assert len(batch0) == B assert len(batch1) == B |
The final output should be the text below, and all the assertions can be passed.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 | mean0 & mean1: 1.3089549541473389 1.308955 even_index_a0: [[[0.68557155 0.0400669 0.94007933 0.78716105 0.20914847 0.8599181 0.92103994 0.47839886] [0.135239 0.68858844 0.6829707 0.58197 0.48305878 0.9397303 0.89459634 0.5577291 ]] [[0.9488533 0.47845793 0.51480126 0.0555109 0.19929752 0.23601894 0.42786413 0.5761327 ] [0.818701 0.28468645 0.25223565 0.43992475 0.81982213 0.07291706 0.71379566 0.91475743]] [[0.9116944 0.24528384 0.07600943 0.47568366 0.84414846 0.10080811 0.48615587 0.42902485] [0.9828625 0.71282893 0.29188994 0.89616627 0.8868826 0.45475304 0.23269841 0.64100367]] [[0.3698722 0.5704599 0.3201713 0.4965511 0.9929494 0.17030907 0.03221584 0.772024 ] [0.9978101 0.13638996 0.19555 0.4724506 0.9358848 0.9002964 0.28848648 0.5132838 ]]] even_index_a1: [[[0.68557155 0.0400669 0.94007933 0.78716105 0.20914847 0.8599181 0.92103994 0.47839886] [0.135239 0.68858844 0.6829707 0.58197 0.48305878 0.9397303 0.89459634 0.5577291 ]] [[0.9488533 0.47845793 0.51480126 0.0555109 0.19929752 0.23601894 0.42786413 0.5761327 ] [0.818701 0.28468645 0.25223565 0.43992475 0.81982213 0.07291706 0.71379566 0.91475743]] [[0.9116944 0.24528384 0.07600943 0.47568366 0.84414846 0.10080811 0.48615587 0.42902485] [0.9828625 0.71282893 0.29188994 0.89616627 0.8868826 0.45475304 0.23269841 0.64100367]] [[0.3698722 0.5704599 0.3201713 0.4965511 0.9929494 0.17030907 0.03221584 0.772024 ] [0.9978101 0.13638996 0.19555 0.4724506 0.9358848 0.9002964 0.28848648 0.5132838 ]]] |
In this case, we can see that the TreeValue
can be properly applied into the numpy
library.
The tree-structured matrix calculation can be easily built with TreeValue
like using standard numpy array.
Both the simplicity of logic structure and execution efficiency can be improve a lot.
And Last but not least, the only thing you need to do is to wrap the functions in Numpy library, and then use it painlessly like the primitive numpy.