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: 1.323505848646164 1.3235059 even_index_a0: [[[0.90539867 0.38841063 0.34561673 0.99831134 0.45578665 0.592899 0.87183154 0.6574434 ] [0.91159683 0.09298854 0.71090615 0.8612857 0.60535073 0.5963975 0.56070673 0.5195303 ]] [[0.23590541 0.84415245 0.8365436 0.665977 0.1597204 0.2570416 0.35903034 0.02168042] [0.74665266 0.47121313 0.88155264 0.29203656 0.22009672 0.82261723 0.9600288 0.2045388 ]] [[0.95245695 0.8706708 0.10267961 0.3616081 0.6888209 0.25009882 0.6990164 0.17128341] [0.510822 0.52112025 0.32132077 0.9448549 0.40553337 0.56547546 0.5634468 0.893396 ]] [[0.23426725 0.3908564 0.3983672 0.06658428 0.2986731 0.7164186 0.24367206 0.660595 ] [0.5391705 0.13134257 0.7809654 0.96425056 0.5202109 0.32863 0.41251257 0.885573 ]]] even_index_a1: [[[0.90539867 0.38841063 0.34561673 0.99831134 0.45578665 0.592899 0.87183154 0.6574434 ] [0.91159683 0.09298854 0.71090615 0.8612857 0.60535073 0.5963975 0.56070673 0.5195303 ]] [[0.23590541 0.84415245 0.8365436 0.665977 0.1597204 0.2570416 0.35903034 0.02168042] [0.74665266 0.47121313 0.88155264 0.29203656 0.22009672 0.82261723 0.9600288 0.2045388 ]] [[0.95245695 0.8706708 0.10267961 0.3616081 0.6888209 0.25009882 0.6990164 0.17128341] [0.510822 0.52112025 0.32132077 0.9448549 0.40553337 0.56547546 0.5634468 0.893396 ]] [[0.23426725 0.3908564 0.3983672 0.06658428 0.2986731 0.7164186 0.24367206 0.660595 ] [0.5391705 0.13134257 0.7809654 0.96425056 0.5202109 0.32863 0.41251257 0.885573 ]]] |
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.