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: 1.2657971382141113 1.2657971 even_index_a0: [[[0.01312 0.48792323 0.3291028 0.00304565 0.6433438 0.74080294 0.5121814 0.5623451 ] [0.5512147 0.39032695 0.35219312 0.504061 0.02856841 0.29818988 0.9053334 0.665941 ]] [[0.9757954 0.68013906 0.10399038 0.5342024 0.37053555 0.5778939 0.41451073 0.909643 ] [0.8712018 0.6481216 0.18073541 0.4383028 0.96272665 0.7050357 0.67057306 0.44109976]] [[0.36717924 0.6180378 0.5249698 0.09944201 0.7087774 0.6306388 0.8574368 0.49080858] [0.7763973 0.08385272 0.52192235 0.74264264 0.33316442 0.56623316 0.63337004 0.29658946]] [[0.65045375 0.04383139 0.8350498 0.8271432 0.03529339 0.9438283 0.24729134 0.7680081 ] [0.6244855 0.91742337 0.42009917 0.108338 0.1630611 0.30331355 0.25903767 0.25184193]]] even_index_a1: [[[0.01312 0.48792323 0.3291028 0.00304565 0.6433438 0.74080294 0.5121814 0.5623451 ] [0.5512147 0.39032695 0.35219312 0.504061 0.02856841 0.29818988 0.9053334 0.665941 ]] [[0.9757954 0.68013906 0.10399038 0.5342024 0.37053555 0.5778939 0.41451073 0.909643 ] [0.8712018 0.6481216 0.18073541 0.4383028 0.96272665 0.7050357 0.67057306 0.44109976]] [[0.36717924 0.6180378 0.5249698 0.09944201 0.7087774 0.6306388 0.8574368 0.49080858] [0.7763973 0.08385272 0.52192235 0.74264264 0.33316442 0.56623316 0.63337004 0.29658946]] [[0.65045375 0.04383139 0.8350498 0.8271432 0.03529339 0.9438283 0.24729134 0.7680081 ] [0.6244855 0.91742337 0.42009917 0.108338 0.1630611 0.30331355 0.25903767 0.25184193]]] |
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.