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: 1.3209200203418732 1.32092 even_index_a0: [[[0.7049845 0.6832885 0.30265245 0.929126 0.6074148 0.44807675 0.62992114 0.83384764] [0.3976229 0.757243 0.63562495 0.44865856 0.8656993 0.2802445 0.99898005 0.70329577]] [[0.33519658 0.77666414 0.26646182 0.920521 0.9298986 0.9654781 0.25486284 0.9555463 ] [0.28609633 0.7333342 0.6247359 0.8514759 0.40342852 0.5305416 0.03246398 0.10308591]] [[0.6460369 0.18840446 0.3076195 0.5131027 0.01366149 0.33337337 0.46806684 0.0473442 ] [0.7950862 0.5373282 0.57397944 0.6824749 0.06575133 0.94041574 0.51668406 0.49237564]] [[0.46737078 0.40297973 0.79974085 0.13291207 0.07290976 0.7090026 0.5181491 0.44200504] [0.19791219 0.16497588 0.46470013 0.71000946 0.3251046 0.7356336 0.06359822 0.6423732 ]]] even_index_a1: [[[0.7049845 0.6832885 0.30265245 0.929126 0.6074148 0.44807675 0.62992114 0.83384764] [0.3976229 0.757243 0.63562495 0.44865856 0.8656993 0.2802445 0.99898005 0.70329577]] [[0.33519658 0.77666414 0.26646182 0.920521 0.9298986 0.9654781 0.25486284 0.9555463 ] [0.28609633 0.7333342 0.6247359 0.8514759 0.40342852 0.5305416 0.03246398 0.10308591]] [[0.6460369 0.18840446 0.3076195 0.5131027 0.01366149 0.33337337 0.46806684 0.0473442 ] [0.7950862 0.5373282 0.57397944 0.6824749 0.06575133 0.94041574 0.51668406 0.49237564]] [[0.46737078 0.40297973 0.79974085 0.13291207 0.07290976 0.7090026 0.5181491 0.44200504] [0.19791219 0.16497588 0.46470013 0.71000946 0.3251046 0.7356336 0.06359822 0.6423732 ]]] |
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.