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: 1.3417839109897614 1.3417839 even_index_a0: [[[0.5405964 0.04800224 0.2011775 0.11673993 0.5924523 0.8629552 0.9306103 0.6450615 ] [0.9261841 0.15643097 0.24029383 0.9997922 0.63324857 0.3303629 0.76098144 0.84606564]] [[0.81188416 0.42624184 0.39276618 0.17766885 0.34539056 0.5035212 0.8416065 0.8475083 ] [0.55965775 0.94468117 0.20583464 0.41816393 0.4608822 0.31241888 0.03827463 0.93020165]] [[0.01444306 0.9404593 0.65871197 0.5354966 0.70934194 0.46803784 0.36677232 0.63359386] [0.18884651 0.07462371 0.25408325 0.00700985 0.25150198 0.24313512 0.6590286 0.25204092]] [[0.37779397 0.6665918 0.62731063 0.39729866 0.6528085 0.58827806 0.03939836 0.4984216 ] [0.18584058 0.36716783 0.3568997 0.38830996 0.60135514 0.8472666 0.7048019 0.4198037 ]]] even_index_a1: [[[0.5405964 0.04800224 0.2011775 0.11673993 0.5924523 0.8629552 0.9306103 0.6450615 ] [0.9261841 0.15643097 0.24029383 0.9997922 0.63324857 0.3303629 0.76098144 0.84606564]] [[0.81188416 0.42624184 0.39276618 0.17766885 0.34539056 0.5035212 0.8416065 0.8475083 ] [0.55965775 0.94468117 0.20583464 0.41816393 0.4608822 0.31241888 0.03827463 0.93020165]] [[0.01444306 0.9404593 0.65871197 0.5354966 0.70934194 0.46803784 0.36677232 0.63359386] [0.18884651 0.07462371 0.25408325 0.00700985 0.25150198 0.24313512 0.6590286 0.25204092]] [[0.37779397 0.6665918 0.62731063 0.39729866 0.6528085 0.58827806 0.03939836 0.4984216 ] [0.18584058 0.36716783 0.3568997 0.38830996 0.60135514 0.8472666 0.7048019 0.4198037 ]]] |
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.