layer_function_generator.py 11.4 KB
Newer Older
1
#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
3 4 5
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
D
dzhwinter 已提交
6 7 8
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
9 10 11 12 13
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
14 15

from __future__ import print_function
D
dzhwinter 已提交
16 17
import re
import functools
18
import warnings
Y
yuyang18 已提交
19
import string
D
dzhwinter 已提交
20

21
from six.moves import cStringIO
22
from ..proto import framework_pb2
W
Wu Yi 已提交
23
from ..framework import OpProtoHolder, Variable, core, convert_np_dtype_to_dtype_
24
from ..layer_helper import LayerHelper
D
dzhwinter 已提交
25

26
__all__ = [
27
    'deprecated', 'generate_layer_fn', 'generate_activation_fn', 'autodoc',
28 29
    'templatedoc'
]
D
dzhwinter 已提交
30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47


def _convert_(name):
    """
    Formatting.

    Args:
       name: The name/alias

    This function takes in a name and converts it to a standard format of
    group1_group2. Where as per the regular expression, group1 can have
    alphabets and numbers and group2 has capital alphabets.

    """
    s1 = re.sub('(.)([A-Z][a-z]+)', r'\1_\2', name)
    return re.sub('([a-z0-9])([A-Z])', r'\1_\2', s1).lower()


Y
yuyang18 已提交
48 49 50 51
def _type_to_str_(tp):
    return framework_pb2.AttrType.Name(tp)


52 53 54 55 56
_two_dollar_pattern_ = re.compile(r"\$\$([^\$]+)\$\$")
_single_dollar_pattern_ = re.compile(r"\$([^\$]+)\$")
_two_bang_pattern_ = re.compile(r"!!([^!]+)!!")


Y
yuyang18 已提交
57 58 59 60 61 62 63
def escape_math(text):
    return _two_bang_pattern_.sub(
        r'$$\1$$',
        _single_dollar_pattern_.sub(r':math:`\1`',
                                    _two_dollar_pattern_.sub(r"!!\1!!", text)))


64 65 66
def _generate_doc_string_(op_proto,
                          additional_args_lines=None,
                          skip_attrs_set=None):
D
dzhwinter 已提交
67 68 69 70 71 72 73 74 75 76 77 78 79
    """
    Generate docstring by OpProto

    Args:
        op_proto (framework_pb2.OpProto): a protobuf message typed OpProto

    Returns:
        str: the document string
    """

    if not isinstance(op_proto, framework_pb2.OpProto):
        raise TypeError("OpProto should be `framework_pb2.OpProto`")

80
    buf = cStringIO()
81
    buf.write(escape_math(op_proto.comment))
D
dzhwinter 已提交
82 83 84 85
    buf.write('\nArgs:\n')
    for each_input in op_proto.inputs:
        line_begin = '    {0}: '.format(_convert_(each_input.name))
        buf.write(line_begin)
86 87 88 89 90
        buf.write(escape_math(each_input.comment))
        if each_input.duplicable:
            buf.write("  Duplicatable.")
        if each_input.dispensable:
            buf.write("  Optional.")
D
dzhwinter 已提交
91 92
        buf.write('\n')

93
    skip_attrs = OpProtoHolder.generated_op_attr_names()
94 95 96
    # attr use_mkldnn and is_test also should not be visible to users.
    skip_attrs.add("use_mkldnn")
    skip_attrs.add("is_test")
97
    skip_attrs.add("use_cudnn")
98 99 100 101 102

    if skip_attrs_set:
        for t in skip_attrs_set:
            skip_attrs.add(t)

D
dzhwinter 已提交
103
    for each_attr in op_proto.attrs:
104 105
        if each_attr.name in skip_attrs:
            continue
D
dzhwinter 已提交
106 107 108 109 110
        buf.write('    ')
        buf.write(each_attr.name)
        buf.write(' (')
        buf.write(_type_to_str_(each_attr.type))
        buf.write('): ')
111
        buf.write(escape_math(each_attr.comment))
D
dzhwinter 已提交
112 113
        buf.write('\n')

S
sneaxiy 已提交
114 115 116 117 118 119 120
    if additional_args_lines is not None:
        for line in additional_args_lines:
            line = line.strip()
            buf.write('    ')
            buf.write(line)
            buf.write('\n')

D
dzhwinter 已提交
121 122 123 124 125 126
    if len(op_proto.outputs) != 0:
        buf.write('\nReturns:\n')
        buf.write('    ')
        for each_opt in op_proto.outputs:
            if not each_opt.intermediate:
                break
127
        buf.write(escape_math(each_opt.comment))
D
dzhwinter 已提交
128 129 130 131

    return buf.getvalue()


132
def generate_layer_fn(op_type):
133
    """Register the Python layer for an Operator.
D
dzhwinter 已提交
134 135

    Args:
136
       op_type: The name of the operator to be created.
D
dzhwinter 已提交
137 138 139 140 141 142 143

    This function takes in the operator type (sigmoid, mean , average etc) and
    creates the operator functionality.

    """
    op_proto = OpProtoHolder.instance().get_op_proto(op_type)
    not_intermediate_outputs = \
144
        [output for output in op_proto.outputs if not output.intermediate]
D
dzhwinter 已提交
145
    intermediate_outputs = \
146
        [output for output in op_proto.outputs if output.intermediate]
D
dzhwinter 已提交
147 148 149

    if len(not_intermediate_outputs) != 1:
        raise ValueError("Only one non intermediate output operator can be",
Y
yuyang18 已提交
150
                         "automatically generated. {0}".format(op_type))
D
dzhwinter 已提交
151 152 153

    if not_intermediate_outputs[0].duplicable:
        raise ValueError(
154
            "Only non duplicable op can be automatically generated.")
D
dzhwinter 已提交
155 156 157 158

    for output in intermediate_outputs:
        if output.duplicable:
            raise ValueError("The op can be automatically generated only when ",
159
                             "all intermediate ops are not duplicable.")
D
dzhwinter 已提交
160 161 162 163

    o_name = not_intermediate_outputs[0].name
    intermediate_output_names = [output.name for output in intermediate_outputs]

Y
Yu Yang 已提交
164
    def infer_and_check_dtype(op_proto, *args, **kwargs):
D
dzhwinter 已提交
165 166 167 168 169 170 171 172 173 174
        """
        This function performs the sanity check for dtype and
        instance type.
        """
        dtype = None
        for ipt in op_proto.inputs:
            name = _convert_(ipt.name)
            val = kwargs.pop(name, [])
            if not isinstance(val, list) and not isinstance(val, tuple):
                val = [val]
Y
Yu Yang 已提交
175 176 177 178
            if len(val) == 0:
                val = [args[0]]
                args = args[1:]

D
dzhwinter 已提交
179 180 181 182 183 184 185 186 187 188 189 190
            for each in val:
                if not isinstance(each, Variable):
                    raise ValueError("input of {0} must be variable".format(
                        op_type))

                if dtype is None:
                    dtype = each.dtype
                elif dtype != each.dtype:
                    raise ValueError(
                        "operator {0} must input same dtype. {1} vs {2}".format(
                            op_type, dtype, each.dtype))

W
Wu Yi 已提交
191 192 193 194 195 196 197 198 199
        if dtype is None:
            arg_dtype = kwargs.get("dtype")
            if arg_dtype:
                if not isinstance(arg_dtype, core.VarDesc.VarType):
                    dtype = convert_np_dtype_to_dtype_(arg_dtype)
                else:
                    dtype = arg_dtype
            else:
                dtype = core.VarDesc.VarType.FP32
D
dzhwinter 已提交
200 201
        return dtype

Y
Yu Yang 已提交
202
    def func(*args, **kwargs):
D
dzhwinter 已提交
203 204
        helper = LayerHelper(op_type, **kwargs)

Y
Yu Yang 已提交
205
        dtype = infer_and_check_dtype(op_proto, *args, **kwargs)
D
dzhwinter 已提交
206 207 208 209 210 211 212

        inputs = dict()
        for ipt in op_proto.inputs:
            name = _convert_(ipt.name)
            val = kwargs.pop(name, [])
            if not isinstance(val, list) and not isinstance(val, tuple):
                val = [val]
Y
Yu Yang 已提交
213 214 215
            if len(val) == 0 and len(args) != 0:
                val = args[0]
                args = args[1:]
D
dzhwinter 已提交
216 217 218
            inputs[ipt.name] = val

        outputs = dict()
219 220 221 222 223
        out = kwargs.pop(_convert_(o_name), [])
        if out:
            out_var = out[0] if (isinstance(out, list) or
                                 isinstance(out, tuple)) else out
        else:
X
Xin Pan 已提交
224
            out_var = helper.create_variable_for_type_inference(dtype=dtype)
225
        outputs[o_name] = [out_var]
D
dzhwinter 已提交
226
        for name in intermediate_output_names:
X
Xin Pan 已提交
227 228 229
            outputs[name] = [
                helper.create_variable_for_type_inference(dtype=dtype)
            ]
D
dzhwinter 已提交
230 231
        helper.append_op(
            type=op_type, inputs=inputs, outputs=outputs, attrs=kwargs)
232
        return helper.append_activation(out_var)
D
dzhwinter 已提交
233 234 235 236 237 238

    func.__name__ = op_type
    func.__doc__ = _generate_doc_string_(op_proto)
    return func


239
def generate_activation_fn(op_type):
240 241 242 243 244 245 246 247 248 249 250
    """Register the Python layer for an Operator without Attribute.

    Args:
       op_type: The name of the operator to be created.

    This function takes in the operator type (sigmoid, exp , tanh etc) and
    creates the operator functionality.

    """
    op_proto = OpProtoHolder.instance().get_op_proto(op_type)

T
tensor-tang 已提交
251
    def func(x, name=None):
252
        helper = LayerHelper(op_type, **locals())
X
Xin Pan 已提交
253
        output = helper.create_variable_for_type_inference(dtype=x.dtype)
T
tensor-tang 已提交
254
        helper.append_op(type=op_type, inputs={"X": x}, outputs={"Out": output})
255 256 257 258
        return output

    func.__name__ = op_type
    func.__doc__ = _generate_doc_string_(op_proto)
259 260 261
    func.__doc__ = func.__doc__ + """
Examples:
    .. code-block:: python
262

263
        import paddle.fluid as fluid
X
xiaoting 已提交
264
        data = fluid.layers.data(name="input", shape=[None, 32, 784])
265 266
        result = fluid.layers.%s(data)
""" % op_type
267 268 269
    return func


D
dzhwinter 已提交
270 271 272 273 274 275 276 277 278 279 280
def deprecated(func_or_class):
    """
    Deprecated warning decorator. It will result a warning message.
    Should be used before class or function, member function
    """

    @functools.wraps(func)
    def func_wrapper(*args, **kwargs):
        """
        Wrap func with deprecated warning
        """
Y
Yang Yu 已提交
281
        warnings.simplefilter('always', DeprecationWarning)  # turn off filter
D
dzhwinter 已提交
282 283 284 285
        warnings.warn(
            "Call to deprecated function {}.".format(func.__name__),
            category=DeprecationWarning,
            stacklevel=2)
Y
Yang Yu 已提交
286
        warnings.simplefilter('default', DeprecationWarning)  # reset filter
D
dzhwinter 已提交
287 288 289
        return func(*args, **kwargs)

    return func_wrapper
Y
Yang Yu 已提交
290 291


292 293 294 295 296 297 298
def autodoc(comment=""):
    def __impl__(func):
        func.__doc__ = _generate_doc_string_(OpProtoHolder.instance(
        ).get_op_proto(func.__name__)) + comment
        return func

    return __impl__
Y
yuyang18 已提交
299 300


Y
yuyang18 已提交
301
def templatedoc(op_type=None):
Y
yuyang18 已提交
302 303 304 305 306 307 308 309 310 311
    """
    Decorator of layer function. It will use the docstring from the layer
    function as the template. The template arguments are:

    * ${comment}: The operator comment written in CPP.
    * ${{name}_comment}: The comment of ${name} written with AddAttr, AddOutput,
        and AddInput. The ${name} is Python snake style. i.e., xxx_xxx.
    * ${{name}_type}: The type of ${name}.

    Returns:
Y
yuyang18 已提交
312
        Decorated function.
Y
yuyang18 已提交
313 314
    """

Y
yuyang18 已提交
315 316 317
    def trim_ending_dot(msg):
        return msg.rstrip('.')

Y
yuyang18 已提交
318
    def __impl__(func):
Y
yuyang18 已提交
319 320 321 322 323
        if op_type is None:
            op_type_name = func.__name__
        else:
            op_type_name = op_type
        op_proto = OpProtoHolder.instance().get_op_proto(op_type_name)
Y
yuyang18 已提交
324
        tmpl = string.Template(func.__doc__)
Y
yuyang18 已提交
325 326 327 328

        comment_lines = op_proto.comment.split("\n")
        comment = ""
        for line in comment_lines:
Y
yuyang18 已提交
329 330
            line = line.strip()
            if len(line) != 0:
Y
yuyang18 已提交
331
                comment += escape_math(line)
Y
yuyang18 已提交
332
                comment += " "
Y
yuyang18 已提交
333 334
            elif len(comment) != 0:
                comment += "\n    \n    "
Y
yuyang18 已提交
335

Y
yuyang18 已提交
336
        args = {"comment": trim_ending_dot(comment)}
Y
yuyang18 已提交
337 338
        for each_input in op_proto.inputs:
            input_name = _convert_(each_input.name)
Y
yuyang18 已提交
339 340
            args["{0}_comment".format(input_name)] = trim_ending_dot(
                each_input.comment)
Y
yuyang18 已提交
341 342 343
            args["{0}_type".format(input_name)] = "Variable"
        for each_attr in op_proto.attrs:
            input_name = _convert_(each_attr.name)
Y
yuyang18 已提交
344 345
            args["{0}_comment".format(input_name)] = trim_ending_dot(
                each_attr.comment)
Y
yuyang18 已提交
346 347 348 349
            args["{0}_type".format(input_name)] = _type_to_str_(each_attr.type)

        for each_opt in op_proto.outputs:
            output_name = _convert_(each_opt.name)
Y
yuyang18 已提交
350 351
            args["{0}_comment".format(output_name)] = trim_ending_dot(
                each_opt.comment)
Y
yuyang18 已提交
352 353 354 355 356
            args["{0}_type".format(output_name)] = "Variable"
        func.__doc__ = tmpl.substitute(args)
        return func

    return __impl__