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: 1.2730734050273895 1.2730733 even_index_a0: [[[0.82554036 0.4176295 0.7646068 0.6790579 0.6258933 0.32710877 0.8264562 0.74017555] [0.5707926 0.4385422 0.0504568 0.8773213 0.83896196 0.33129868 0.19768815 0.49905354]] [[0.47267896 0.99706334 0.80905867 0.05866001 0.09228715 0.37031966 0.36674663 0.31362286] [0.09579414 0.96156794 0.75059223 0.6832822 0.38935086 0.454897 0.92272246 0.8728762 ]] [[0.42595568 0.63587976 0.9852289 0.45401752 0.32028416 0.5594362 0.10364991 0.5973567 ] [0.2060748 0.15850422 0.8340456 0.25474724 0.26743057 0.56029 0.16379364 0.13809139]] [[0.91421515 0.696699 0.95589566 0.19583474 0.660949 0.14174311 0.10829286 0.07636495] [0.94169277 0.38197067 0.16279665 0.44453254 0.7807179 0.96516705 0.5172212 0.5586753 ]]] even_index_a1: [[[0.82554036 0.4176295 0.7646068 0.6790579 0.6258933 0.32710877 0.8264562 0.74017555] [0.5707926 0.4385422 0.0504568 0.8773213 0.83896196 0.33129868 0.19768815 0.49905354]] [[0.47267896 0.99706334 0.80905867 0.05866001 0.09228715 0.37031966 0.36674663 0.31362286] [0.09579414 0.96156794 0.75059223 0.6832822 0.38935086 0.454897 0.92272246 0.8728762 ]] [[0.42595568 0.63587976 0.9852289 0.45401752 0.32028416 0.5594362 0.10364991 0.5973567 ] [0.2060748 0.15850422 0.8340456 0.25474724 0.26743057 0.56029 0.16379364 0.13809139]] [[0.91421515 0.696699 0.95589566 0.19583474 0.660949 0.14174311 0.10829286 0.07636495] [0.94169277 0.38197067 0.16279665 0.44453254 0.7807179 0.96516705 0.5172212 0.5586753 ]]] |
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.