io.py 17.1 KB
Newer Older
1
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
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
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13 14
# 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.

Y
Yu Yang 已提交
15
from .. import core
T
typhoonzero 已提交
16
from ..framework import convert_np_dtype_to_dtype_, default_main_program, default_startup_program, Program
Y
Yu Yang 已提交
17
from ..unique_name import generate as unique_name
T
WIP  
typhoonzero 已提交
18 19
from control_flow import BlockGuard
from ..layer_helper import LayerHelper
Y
Refine  
Yu Yang 已提交
20
from ..executor import global_scope
Y
Yu Yang 已提交
21

Y
Yu Yang 已提交
22 23
__all__ = [
    'data', 'BlockGuardServ', 'ListenAndServ', 'Send', 'open_recordio_file',
J
JiayiFeng 已提交
24
    'open_files', 'read_file', 'shuffle', 'batch', 'double_buffer'
Y
Yu Yang 已提交
25
]
Y
Yu Yang 已提交
26 27 28 29 30 31 32 33 34 35


def data(name,
         shape,
         append_batch_size=True,
         dtype='float32',
         lod_level=0,
         type=core.VarDesc.VarType.LOD_TENSOR,
         stop_gradient=True):
    """
K
kavyasrinet 已提交
36
    **Data Layer**
Y
Yu Yang 已提交
37

K
kavyasrinet 已提交
38
    This function takes in the input and based on whether data has
C
caoying03 已提交
39
    to be returned back as a minibatch, it creates the global variable by using
Y
Yu Yang 已提交
40
    the helper functions. The global variables can be accessed by all the
C
caoying03 已提交
41
    following operators in the graph.
Y
Yu Yang 已提交
42 43 44 45

    All the input variables of this function are passed in as local variables
    to the LayerHelper constructor.

K
kavyasrinet 已提交
46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
    Args:
       name(str): The name/alias of the function
       shape(list): Tuple declaring the shape.
       append_batch_size(bool): Whether or not to append the data as a batch.
       dtype(int|float): The type of data : float32, float_16, int etc
       type(VarType): The output type. By default it is LOD_TENSOR.
       lod_level(int): The LoD Level. 0 means the input data is not a sequence.
       main_program(Program): Name of the main program that calls this
       startup_program(Program): Name of the startup program
       stop_gradient(bool): A boolean that mentions whether gradient should flow.

    Returns:
        Variable: The global variable that gives access to the data.

    Examples:
        .. code-block:: python

          data = fluid.layers.data(name='x', shape=[784], dtype='float32')
Y
Yu Yang 已提交
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
    """
    helper = LayerHelper('data', **locals())
    shape = list(shape)
    for i in xrange(len(shape)):
        if shape[i] is None:
            shape[i] = -1
            append_batch_size = False
        elif shape[i] < 0:
            append_batch_size = False

    if append_batch_size:
        shape = [-1] + shape  # append batch size as -1

    return helper.create_global_variable(
        name=name,
        shape=shape,
        dtype=dtype,
        type=type,
        stop_gradient=stop_gradient,
        lod_level=lod_level)
T
typhoonzero 已提交
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114


class BlockGuardServ(BlockGuard):
    """
    BlockGuardServ class.

    BlockGuardServ class is used to create an op with a block in a program.
    """

    def __init__(self, server):
        if not (isinstance(server, ListenAndServ)):
            raise TypeError("BlockGuardServ takes a ListenAndServ")
        super(BlockGuardServ, self).__init__(server.helper.main_program)
        self.server = server

    def __exit__(self, exc_type, exc_val, exc_tb):
        if exc_type is not None:
            return False

        self.server.complete_op()
        return super(BlockGuardServ, self).__exit__(exc_type, exc_val, exc_tb)


class ListenAndServ(object):
    """
    ListenAndServ class.

    ListenAndServ class is used to wrap listen_and_serv op to create a server
    which can receive variables from clients and run a block.
    """

Y
Yancey1989 已提交
115
    def __init__(self, endpoint, inputs, fan_in=1, optimizer_mode=True):
116
        self.helper = LayerHelper("listen_and_serv")
Y
Yancey1989 已提交
117
        self.inputs = inputs
T
typhoonzero 已提交
118 119 120
        self.outputs = []
        self.endpoint = endpoint
        self.fan_in = fan_in
T
typhoonzero 已提交
121 122
        # FIXME(typhoonzero): add optimizer_mode is stupid, should make it more
        # general.
T
WIP  
typhoonzero 已提交
123
        self.optimizer_mode = optimizer_mode
T
typhoonzero 已提交
124 125 126 127 128 129 130 131 132 133 134 135 136

    def do(self):
        return BlockGuardServ(self)

    def get_params_and_grads(self):
        main_program = self.helper.main_program
        current_block = main_program.current_block()
        parent_block = self.parent_block()
        # params and grads in the same order.
        params = list()
        grads = list()
        for op in current_block.ops:
            # FIXME(typhoonzero): op.inputs is None if it's cloned.
T
WIP  
typhoonzero 已提交
137 138 139 140 141 142 143 144
            if self.optimizer_mode:
                if "Grad" in op.inputs and "Param" in op.inputs:
                    params.append(op.inputs["Param"].name)
                    grads.append(op.inputs["Grad"].name)
            else:
                # simple recv mode, recv operators inputs.
                for iname in op.input_names:
                    for in_var_name in op.input(iname):
T
typhoonzero 已提交
145 146
                        params.append(parent_block.var(in_var_name))
                        grads.append(parent_block.var(in_var_name))
T
typhoonzero 已提交
147 148 149

        return params, grads

T
typhoonzero 已提交
150 151 152 153 154 155 156
    def parent_block(self):
        prog = self.helper.main_program
        parent_idx = prog.current_block().parent_idx
        assert parent_idx >= 0
        parent_block = prog.block(parent_idx)
        return parent_block

T
typhoonzero 已提交
157 158 159 160
    def complete_op(self):
        main_program = self.helper.main_program
        current_block = main_program.current_block()
        parent_block = self.parent_block()
T
typhoonzero 已提交
161
        empty_block = Program().global_block()
T
typhoonzero 已提交
162 163

        parent_block.append_op(
164
            type='listen_and_serv',
Y
Yancey1989 已提交
165
            inputs={"X": self.inputs},
T
typhoonzero 已提交
166 167 168 169
            outputs={},
            attrs={
                'endpoint': self.endpoint,
                'Fanin': self.fan_in,
T
typhoonzero 已提交
170
                'OptimizeBlock': current_block,
171 172
                'PrefetchBlock': empty_block,
                'sync_mode': True,  # did not support async now in layers
Q
qiaolongfei 已提交
173
                'grad_to_block_id': [""]
T
typhoonzero 已提交
174 175 176
            })


T
typhoonzero 已提交
177
def Send(endpoints, send_vars, get_vars=None):
T
typhoonzero 已提交
178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
    """
    Send layer

    Args:
        endpoints: comma seperated IP:PORT pairs in the order
                   of send_vars to send
        send_vars: vars to send
        get_vars: vars to get from server after send completes.

    Send variables to the server side, and get vars from server
    side when server have finished running server side program.
    """
    assert (type(send_vars) == list)

    epmap = endpoints.split(",")
T
typhoonzero 已提交
193
    endpoints = list(set(epmap))
T
typhoonzero 已提交
194 195

    helper = LayerHelper("Send", **locals())
Y
Yancey1989 已提交
196 197
    rpc_client_var = default_main_program().global_block().create_var(
        name="RPC_CLIENT_VAR", persistable=True, type=core.VarDesc.VarType.RAW)
T
typhoonzero 已提交
198 199 200 201 202
    if not get_vars:
        get_vars = []
        for s in send_vars:
            v = helper.create_tmp_variable(dtype=s.dtype, stop_gradient=True)
            get_vars.append(v)
Y
Yancey1989 已提交
203

T
typhoonzero 已提交
204 205 206
    helper.append_op(
        type="send",
        inputs={"X": send_vars},
Y
Yancey1989 已提交
207 208
        outputs={"Out": get_vars,
                 "RPCClient": rpc_client_var},
T
typhoonzero 已提交
209 210
        attrs={"endpoints": endpoints,
               "epmap": epmap})
T
typhoonzero 已提交
211
    return get_vars
212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239


def Recv(endpoints, get_vars):
    """
    Recv layer

    Args:
        endpoints: comma seperated IP:PORT pairs in the order
                   of send_vars to send
        send_vars: vars to send
        get_vars: vars to get from server after send completes.

    Send variables to the server side, and get vars from server
    side when server have finished running server side program.
    """
    assert (type(send_vars) == list)
    assert (type(get_vars) == list)

    epmap = endpoints.split(",")
    endpoints = list(set(epmap))

    helper = LayerHelper("Recv", **locals())
    helper.append_op(
        type="recv",
        inputs={"X": get_vars},
        outputs={"Out": get_vars},
        attrs={"endpoints": endpoints,
               "epmap": epmap})
Y
Yu Yang 已提交
240 241


Y
Refine  
Yu Yang 已提交
242 243 244 245 246 247 248 249 250 251
def monkey_patch_reader_methods(reader):
    def __get_reader__():
        scope = global_scope()
        var = scope.find_var(reader.name)
        return var.get_reader()

    def reset():
        return __get_reader__().reset()

    reader.reset = reset
Y
Yu Yang 已提交
252 253
    reader.stop_gradient = True
    reader.persistable = True
Y
Refine  
Yu Yang 已提交
254 255 256
    return reader


Y
Yu Yang 已提交
257 258 259 260 261
def _copy_reader_var_(block, var):
    new_var = block.create_var(name=var.name, type=core.VarDesc.VarType.READER)
    new_var.desc.set_shapes(var.desc.shapes())
    new_var.desc.set_dtypes(var.desc.dtypes())
    new_var.persistable = True
F
fengjiayi 已提交
262 263 264 265
    return new_var


def _copy_reader_create_op_(block, op):
F
fengjiayi 已提交
266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281
    input_param_names = op.input_names
    new_input_map = {}
    for param_name in input_param_names:
        new_input_map[param_name] = []
        arg_names = op.input(param_name)
        for arg_name in arg_names:
            new_input_map[param_name].append(block.var(arg_name))

    output_param_names = op.output_names
    new_output_map = {}
    for param_name in output_param_names:
        new_output_map[param_name] = []
        arg_names = op.output(param_name)
        for arg_name in arg_names:
            new_output_map[param_name].append(block.var(arg_name))

F
fengjiayi 已提交
282
    new_op = block.append_op(
F
fengjiayi 已提交
283 284 285
        type=op.type,
        inputs=new_input_map,
        outputs=new_output_map,
J
JiayiFeng 已提交
286
        attrs=op.all_attrs())
F
fengjiayi 已提交
287
    return new_op
Y
Yu Yang 已提交
288 289


F
fengjiayi 已提交
290 291 292 293 294
def open_recordio_file(filename,
                       shapes,
                       lod_levels,
                       dtypes,
                       pass_num=1,
F
fengjiayi 已提交
295
                       for_parallel=True):
F
fengjiayi 已提交
296 297 298 299 300 301 302 303 304 305 306
    """
    Open a RecordIO file

    This layer takes a RecordIO file to read from and returns a Reader Variable.
    Via the Reader Variable, we can get data from the given RecordIO file.

    Args:
       filename(str): The RecordIO file's name.
       shapes(list): List of tuples which declaring data shapes.
       lod_levels(list): List of ints which declaring data lod_level.
       dtypes(list): List of strs which declaring data type.
F
fengjiayi 已提交
307
       pass_num(int): Number of passes to run.
F
fengjiayi 已提交
308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
       for_parallel(Bool): Set it as True if you are going to run
            subsequent operators in parallel.

    Returns:
       Variable: A Reader Variable via which we can get RecordIO file data.

    Examples:
       .. code-block:: python

         reader = fluid.layers.io.open_recordio_file(
                                          filename='./data.recordio',
                                          shapes=[(3,224,224), (1)],
                                          lod_levels=[0, 0],
                                          dtypes=['float32', 'int64'])

         # Via the reader, we can use 'read_file' layer to get data:
         image, label = fluid.layers.read_file(reader)
    """
Y
Yu Yang 已提交
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349
    dtypes = [convert_np_dtype_to_dtype_(dt) for dt in dtypes]
    shape_concat = []
    ranks = []

    for shape in shapes:
        shape_concat.extend(shape)
        ranks.append(len(shape))

    var_name = unique_name('open_recordio_file')

    startup_blk = default_startup_program().current_block()
    startup_var = startup_blk.create_var(name=var_name)
    startup_blk.append_op(
        type='create_recordio_file_reader',
        outputs={'Out': [startup_var]},
        attrs={
            'shape_concat': shape_concat,
            'lod_levels': lod_levels,
            'filename': filename,
            'ranks': ranks
        })

    startup_var.desc.set_dtypes(dtypes)
    startup_var.persistable = True
F
fengjiayi 已提交
350 351
    main_prog_var = _copy_reader_var_(default_main_program().current_block(),
                                      startup_var)
F
fengjiayi 已提交
352 353 354 355 356

    if pass_num > 1:
        main_prog_var = multi_pass(reader=main_prog_var, pass_num=pass_num)

    if for_parallel:
J
JiayiFeng 已提交
357
        main_prog_var = parallel(reader=main_prog_var)
F
fengjiayi 已提交
358

F
fengjiayi 已提交
359
    return monkey_patch_reader_methods(main_prog_var)
Y
Yu Yang 已提交
360 361


362 363 364 365 366
def open_files(filenames,
               shapes,
               lod_levels,
               dtypes,
               thread_num,
F
fengjiayi 已提交
367 368
               buffer_size=None,
               pass_num=1,
F
fengjiayi 已提交
369
               for_parallel=True):
F
fengjiayi 已提交
370 371 372
    """
    Open files

F
fengjiayi 已提交
373 374 375
    This layer takes a list of files to read from and returns a Reader Variable. 
    Via the Reader Variable, we can get data from given files. All files must 
    have name suffixs to indicate their formats, e.g., '*.recordio'. 
F
fengjiayi 已提交
376 377 378 379 380 381 382 383

    Args:
       filenames(list): The list of file names.
       shapes(list): List of tuples which declaring data shapes.
       lod_levels(list): List of ints which declaring data lod_level.
       dtypes(list): List of strs which declaring data type.
       thread_num(int): The maximal concurrent prefetch thread number.
       buffer_size(int): The size of prefetch buffer.
F
fengjiayi 已提交
384
       pass_num(int): Number of passes to run.
F
fengjiayi 已提交
385 386
       for_parallel(Bool): Set it as True if you are going to run 
            subsequent operators in parallel.
F
fengjiayi 已提交
387 388 389 390 391 392 393

    Returns:
       Variable: A Reader Variable via which we can get file data.

    Examples:
       .. code-block:: python

F
fengjiayi 已提交
394
         reader = fluid.layers.io.open_files(filenames=['./data1.recordio',
F
fengjiayi 已提交
395
                                                     './data2.recordio'],
F
fengjiayi 已提交
396 397 398 399 400
                                             shapes=[(3,224,224), (1)],
                                             lod_levels=[0, 0],
                                             dtypes=['float32', 'int64'],
                                             thread_num=2,
                                             buffer_size=2)
F
fengjiayi 已提交
401 402

         # Via the reader, we can use 'read_file' layer to get data:
F
fengjiayi 已提交
403
         image, label = fluid.layers.io.read_file(reader)
F
fengjiayi 已提交
404
    """
405 406
    if buffer_size is None:
        buffer_size = thread_num
F
fengjiayi 已提交
407 408
    if isinstance(filenames, basestring):
        filenames = [filenames]
F
fengjiayi 已提交
409 410 411 412 413 414 415 416
    dtypes = [convert_np_dtype_to_dtype_(dt) for dt in dtypes]
    shape_concat = []
    ranks = []

    for shape in shapes:
        shape_concat.extend(shape)
        ranks.append(len(shape))

F
fengjiayi 已提交
417
    multi_file_reader_name = unique_name('multi_file_reader')
F
fengjiayi 已提交
418
    startup_blk = default_startup_program().current_block()
F
fengjiayi 已提交
419
    startup_reader = startup_blk.create_var(name=multi_file_reader_name)
F
fengjiayi 已提交
420 421
    startup_blk.append_op(
        type='open_files',
F
fengjiayi 已提交
422
        outputs={'Out': [startup_reader]},
F
fengjiayi 已提交
423 424 425 426
        attrs={
            'shape_concat': shape_concat,
            'lod_levels': lod_levels,
            'ranks': ranks,
F
fengjiayi 已提交
427
            'file_names': filenames,
428 429
            'thread_num': thread_num,
            'buffer_size': buffer_size
F
fengjiayi 已提交
430 431
        })

F
fengjiayi 已提交
432 433 434 435 436 437 438
    startup_reader.desc.set_dtypes(dtypes)
    startup_reader.persistable = True
    main_prog_reader = _copy_reader_var_(default_main_program().current_block(),
                                         startup_reader)
    if pass_num > 1:
        main_prog_reader = multi_pass(
            reader=main_prog_reader, pass_num=pass_num)
F
fengjiayi 已提交
439

F
fengjiayi 已提交
440
    if for_parallel:
J
JiayiFeng 已提交
441
        main_prog_reader = parallel(reader=main_prog_reader)
F
fengjiayi 已提交
442

F
fengjiayi 已提交
443 444 445
    return monkey_patch_reader_methods(main_prog_reader)


J
JiayiFeng 已提交
446
def __create_shared_decorated_reader__(op_type, reader, attrs):
Y
Yu Yang 已提交
447 448 449
    var_name = unique_name(op_type)
    startup_blk = default_startup_program().current_block()
    startup_var = startup_blk.create_var(name=var_name)
F
fengjiayi 已提交
450
    startop_op = startup_blk.append_op(
Y
Yu Yang 已提交
451 452 453 454 455
        type=op_type,
        inputs={'UnderlyingReader': reader},
        outputs={'Out': [startup_var]},
        attrs=attrs)
    startup_var.persistable = True
F
fengjiayi 已提交
456 457 458 459
    main_prog_block = default_main_program().current_block()
    main_prog_var = _copy_reader_var_(main_prog_block, startup_var)
    _copy_reader_create_op_(main_prog_block, startop_op)
    return monkey_patch_reader_methods(main_prog_var)
Y
Yu Yang 已提交
460 461


462 463
def __create_unshared_decorated_reader__(op_type, reader, attrs, name=None):
    new_reader_name = name if name is not None else unique_name(op_type)
464 465 466 467 468 469 470 471 472 473 474 475
    main_blk = default_main_program().current_block()
    new_reader = main_blk.create_var(name=new_reader_name)
    main_blk.append_op(
        type=op_type,
        inputs={'UnderlyingReader': reader},
        outputs={'Out': [new_reader]},
        attrs=attrs)
    new_reader.persistable = True
    new_reader.stop_gradient = True
    return monkey_patch_reader_methods(new_reader)


F
fengjiayi 已提交
476
def shuffle(reader, buffer_size):
477 478
    return __create_unshared_decorated_reader__(
        'create_shuffle_reader', reader, {'buffer_size': int(buffer_size)})
Y
Yu Yang 已提交
479 480


J
JiayiFeng 已提交
481 482 483 484 485
def batch(reader, batch_size):
    return __create_unshared_decorated_reader__(
        'create_batch_reader', reader, {'batch_size': int(batch_size)})


486
def double_buffer(reader, place=None, name=None):
Y
Yu Yang 已提交
487 488 489
    attrs = dict()
    if place is not None:
        attrs['place'] = str(place).upper()
490 491
    return __create_unshared_decorated_reader__(
        'create_double_buffer_reader', reader, attrs, name=name)
Y
Yu Yang 已提交
492 493


F
fengjiayi 已提交
494
def multi_pass(reader, pass_num):
495 496
    return __create_shared_decorated_reader__(
        'create_multi_pass_reader', reader, {'pass_num': int(pass_num)})
F
fengjiayi 已提交
497 498


J
JiayiFeng 已提交
499
def parallel(reader):
J
JiayiFeng 已提交
500 501
    return __create_shared_decorated_reader__('create_threaded_reader', reader,
                                              {})
F
fengjiayi 已提交
502 503


Y
Yu Yang 已提交
504 505 506 507 508
def read_file(file_obj):
    helper = LayerHelper('read_file')
    out = [
        helper.create_tmp_variable(
            stop_gradient=True, dtype='float32')
Y
Yu Yang 已提交
509
        for _ in range(len(file_obj.desc.shapes()))
Y
Yu Yang 已提交
510 511 512 513 514 515 516
    ]
    helper.append_op(
        type='read', inputs={'Reader': [file_obj]}, outputs={'Out': out})
    if len(out) == 1:
        return out[0]
    else:
        return out