Jeremy Johnson | be1a940 | 2021-12-15 17:14:56 +0000 | [diff] [blame] | 1 | """Conversion utility from binary numpy files to JSON and the reverse.""" |
| 2 | # Copyright (c) 2021-2022, ARM Limited. |
| 3 | # SPDX-License-Identifier: Apache-2.0 |
| 4 | import json |
| 5 | from pathlib import Path |
| 6 | from typing import Optional |
| 7 | from typing import Union |
| 8 | |
| 9 | import numpy as np |
| 10 | |
| 11 | |
| 12 | class NumpyArrayEncoder(json.JSONEncoder): |
| 13 | """A JSON encoder for Numpy data types.""" |
| 14 | |
| 15 | def default(self, obj): |
| 16 | """Encode default operation.""" |
| 17 | if isinstance(obj, np.integer): |
| 18 | return int(obj) |
James Ward | 8b39043 | 2022-08-12 20:48:56 +0100 | [diff] [blame] | 19 | elif isinstance(obj, np.float32): |
Jeremy Johnson | be1a940 | 2021-12-15 17:14:56 +0000 | [diff] [blame] | 20 | return float(obj) |
James Ward | 8b39043 | 2022-08-12 20:48:56 +0100 | [diff] [blame] | 21 | elif isinstance(obj, np.float16): |
| 22 | return np.float16(obj) |
Jeremy Johnson | be1a940 | 2021-12-15 17:14:56 +0000 | [diff] [blame] | 23 | elif isinstance(obj, np.ndarray): |
| 24 | return obj.tolist() |
| 25 | return super(NumpyArrayEncoder, self).default(obj) |
| 26 | |
| 27 | |
| 28 | def get_shape(t: Union[list, tuple]): |
| 29 | """Get the shape of an N-Dimensional tensor.""" |
| 30 | # TODO: validate shape is consistent for all rows and ccolumns |
| 31 | if isinstance(t, (list, tuple)) and t: |
| 32 | return [len(t)] + get_shape(t[0]) |
| 33 | return [] |
| 34 | |
| 35 | |
| 36 | def npy_to_json(n_path: Path, j_path: Optional[Path] = None): |
| 37 | """Load a numpy data file and save it as a JSON file. |
| 38 | |
| 39 | n_path: the Path to the numpy file |
| 40 | j_path: the Path to the JSON file, if None, it is derived from n_path |
| 41 | """ |
| 42 | if not j_path: |
| 43 | j_path = n_path.parent / (n_path.stem + ".json") |
| 44 | with open(n_path, "rb") as fd: |
| 45 | data = np.load(fd) |
| 46 | jdata = { |
| 47 | "type": data.dtype.name, |
| 48 | "data": data.tolist(), |
| 49 | } |
| 50 | with open(j_path, "w") as fp: |
| 51 | json.dump(jdata, fp, indent=2) |
| 52 | |
| 53 | |
| 54 | def json_to_npy(j_path: Path, n_path: Optional[Path] = None): |
| 55 | """Load a JSON file and save it as a numpy data file. |
| 56 | |
| 57 | j_path: the Path to the JSON file |
| 58 | n_path: the Path to the numpy file, if None, it is derived from j_path |
| 59 | """ |
| 60 | if not n_path: |
| 61 | n_path = j_path.parent / (j_path.stem + ".npy") |
| 62 | with open(j_path, "rb") as fd: |
| 63 | jdata = json.load(fd) |
| 64 | raw_data = jdata["data"] |
| 65 | raw_type = jdata["type"] |
| 66 | shape = get_shape(raw_data) |
| 67 | data = np.asarray(raw_data).reshape(shape).astype(raw_type) |
| 68 | with open(n_path, "wb") as fd: |
| 69 | np.save(fd, data) |
| 70 | |
| 71 | |
| 72 | # ------------------------------------------------------------------------------ |
| 73 | |
| 74 | |
| 75 | def test(): |
| 76 | """Test conversion routines.""" |
| 77 | shape = [2, 3, 4] |
| 78 | elements = 1 |
| 79 | for i in shape: |
| 80 | elements *= i |
| 81 | |
| 82 | # file names |
| 83 | n_path = Path("data.npy") |
| 84 | j_path = Path("data.json") |
| 85 | j2n_path = Path("data_j2n.npy") |
| 86 | |
| 87 | datatypes = [ |
| 88 | np.bool_, |
| 89 | np.int8, |
| 90 | np.int16, |
| 91 | np.int32, |
| 92 | np.int64, |
| 93 | np.uint8, |
| 94 | np.uint16, |
| 95 | np.uint32, |
| 96 | np.uint64, |
| 97 | np.float16, |
| 98 | np.float32, |
| 99 | np.float64, |
| 100 | # np.float128, |
| 101 | # np.complex64, |
| 102 | # np.complex128, |
| 103 | # np.complex256, |
| 104 | # np.datetime64, |
| 105 | # np.str, |
| 106 | ] |
| 107 | |
| 108 | for data_type in datatypes: |
| 109 | dt = np.dtype(data_type) |
| 110 | print(data_type, dt, dt.char, dt.num, dt.name, dt.str) |
| 111 | |
| 112 | # create a tensor of the given shape |
| 113 | tensor = np.arange(elements).reshape(shape).astype(data_type) |
| 114 | # print(tensor) |
| 115 | |
| 116 | # save the tensor in a binary numpy file |
| 117 | with open(n_path, "wb") as fd: |
| 118 | np.save(fd, tensor) |
| 119 | |
| 120 | # read back the numpy file for verification |
| 121 | with open(n_path, "rb") as fd: |
| 122 | tensor1 = np.load(fd) |
| 123 | |
| 124 | # confirm the loaded tensor matches the original |
| 125 | assert tensor.shape == tensor1.shape |
| 126 | assert tensor.dtype == tensor1.dtype |
| 127 | assert (tensor == tensor1).all() |
| 128 | |
| 129 | # convert the numpy file to json |
| 130 | npy_to_json(n_path, j_path) |
| 131 | |
| 132 | # convert the json file to numpy |
| 133 | json_to_npy(j_path, j2n_path) |
| 134 | |
| 135 | # read back the json-to-numpy file for verification |
| 136 | with open(j2n_path, "rb") as fd: |
| 137 | tensor1 = np.load(fd) |
| 138 | |
| 139 | # confirm the loaded tensor matches the original |
| 140 | assert tensor.shape == tensor1.shape |
| 141 | assert tensor.dtype == tensor1.dtype |
| 142 | assert (tensor == tensor1).all() |
| 143 | |
| 144 | # delete the files, if no problems were found |
| 145 | # they are left for debugging if any of the asserts failed |
| 146 | n_path.unlink() |
| 147 | j_path.unlink() |
| 148 | j2n_path.unlink() |
| 149 | return 0 |
| 150 | |
| 151 | |
| 152 | def main(argv=None): |
| 153 | """Load and convert supplied file based on file suffix.""" |
| 154 | import argparse |
| 155 | |
| 156 | parser = argparse.ArgumentParser() |
| 157 | parser.add_argument( |
| 158 | "path", type=Path, help="the path to the file to convert, or 'test'" |
| 159 | ) |
| 160 | args = parser.parse_args(argv) |
| 161 | path = args.path |
| 162 | if str(path) == "test": |
| 163 | print("test") |
| 164 | return test() |
| 165 | |
| 166 | if not path.is_file(): |
| 167 | print(f"Invalid file - {path}") |
| 168 | return 2 |
| 169 | |
| 170 | if path.suffix == ".npy": |
| 171 | npy_to_json(path) |
| 172 | elif path.suffix == ".json": |
| 173 | json_to_npy(path) |
| 174 | else: |
| 175 | print("Unknown file type - {path.suffix}") |
| 176 | return 2 |
| 177 | |
| 178 | return 0 |
| 179 | |
| 180 | |
| 181 | if __name__ == "__main__": |
| 182 | exit(main()) |