Source code for mxnet.module.sequential_module

# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you 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
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# 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.

# pylint: disable=too-many-arguments, too-many-locals, too-many-instance-attributes
"""`SequentialModule` is a container module that chains a number of modules together."""

import logging
import copy

from ..initializer import Uniform

from .base_module import BaseModule

[docs]class SequentialModule(BaseModule): """A SequentialModule is a container module that can chain multiple modules together. .. note:: Building a computation graph with this kind of imperative container is less flexible and less efficient than the symbolic graph. So, this should be only used as a handy utility. """ META_TAKE_LABELS = 'take_labels' META_AUTO_WIRING = 'auto_wiring' def __init__(self, logger=logging): super(SequentialModule, self).__init__(logger=logger) self._modules = [] self._metas = [] self._label_shapes = None self._data_shapes = None self._meta_keys = set([getattr(SequentialModule, x) for x in dir(SequentialModule) if x.startswith('META_')])
[docs] def add(self, module, **kwargs): """Adds a module to the chain. Parameters ---------- module : BaseModule The new module to add. kwargs : **keywords All the keyword arguments are saved as meta information for the added module. The currently known meta includes - `take_labels`: indicating whether the module expect to take labels when doing computation. Note any module in the chain can take labels (not necessarily only the top most one), and they all take the same labels passed from the original data batch for the `SequentialModule`. Returns ------- self This function returns `self` to allow us to easily chain a series of `add` calls. Examples -------- >>> # An example of addinging two modules to a chain. >>> seq_mod = mx.mod.SequentialModule() >>> seq_mod.add(mod1) >>> seq_mod.add(mod2) """ self._modules.append(module) # a sanity check to avoid typo for key in kwargs: assert key in self._meta_keys, ('Unknown meta "%s", a typo?' % key) self._metas.append(kwargs) # after adding new modules, we are reset back to raw states, needs # to bind, init_params, etc. self.binded = False self.params_initialized = False self.optimizer_initialized = False return self # for easier chaining
@property def data_names(self): """A list of names for data required by this module.""" if len(self._modules) > 0: return self._modules[0].data_names return [] @property def output_names(self): """A list of names for the outputs of this module.""" if len(self._modules) > 0: return self._modules[-1].output_names return [] @property def data_shapes(self): """Gets data shapes. Returns ------- list A list of `(name, shape)` pairs. The data shapes of the first module is the data shape of a `SequentialModule`. """ assert self.binded return self._modules[0].data_shapes @property def label_shapes(self): """Gets label shapes. Returns ------- list A list of `(name, shape)` pairs. The return value could be `None` if the module does not need labels, or if the module is not bound for training (in this case, label information is not available). """ assert self.binded return self._label_shapes @property def output_shapes(self): """Gets output shapes. Returns ------- list A list of `(name, shape)` pairs. The output shapes of the last module is the output shape of a `SequentialModule`. """ assert self.binded return self._modules[-1].output_shapes
[docs] def get_params(self): """Gets current parameters. Returns ------- (arg_params, aux_params) A pair of dictionaries each mapping parameter names to NDArray values. This is a merged dictionary of all the parameters in the modules. """ assert self.binded and self.params_initialized arg_params = dict() aux_params = dict() for module in self._modules: arg, aux = module.get_params() arg_params.update(arg) aux_params.update(aux) return (arg_params, aux_params)
[docs] def init_params(self, initializer=Uniform(0.01), arg_params=None, aux_params=None, allow_missing=False, force_init=False, allow_extra=False): """Initializes parameters. Parameters ---------- initializer : Initializer arg_params : dict Default ``None``. Existing parameters. This has higher priority than `initializer`. aux_params : dict Default ``None``. Existing auxiliary states. This has higher priority than `initializer`. allow_missing : bool Allow missing values in `arg_params` and `aux_params` (if not ``None``). In this case, missing values will be filled with `initializer`. force_init : bool Default ``False``. allow_extra : boolean, optional Whether allow extra parameters that are not needed by symbol. If this is True, no error will be thrown when arg_params or aux_params contain extra parameters that is not needed by the executor. """ if self.params_initialized and not force_init: return assert self.binded, 'call bind before initializing the parameters' for module in self._modules: module.init_params(initializer=initializer, arg_params=arg_params, aux_params=aux_params, allow_missing=allow_missing, force_init=force_init, allow_extra=allow_extra) # make sure we do not have duplicated parameter names def _check_name(known_names, new_names, modules, i): """Internal function to help checking duplicated names.""" for name in new_names: assert not name in known_names, "Duplicated parameter names: " + \ ('name "%s" in layer %d (%s) is already ' % (name, i, type(modules[i]))) + \ ('used in layer %d (%s).' % (known_names[name], type(modules[known_names[name]]))) known_names[name] = i arg_names = dict() aux_names = dict() for i_layer, module in enumerate(self._modules): arg_params, aux_params = module.get_params() _check_name(arg_names, arg_params.keys(), self._modules, i_layer) _check_name(aux_names, aux_params.keys(), self._modules, i_layer) self.params_initialized = True
[docs] def bind(self, data_shapes, label_shapes=None, for_training=True, inputs_need_grad=False, force_rebind=False, shared_module=None, grad_req='write'): """Binds the symbols to construct executors. This is necessary before one can perform computation with the module. Parameters ---------- data_shapes : list of (str, tuple) Typically is `data_iter.provide_data`. label_shapes : list of (str, tuple) Typically is `data_iter.provide_label`. for_training : bool Default is ``True``. Whether the executors should be bind for training. inputs_need_grad : bool Default is ``False``. Whether the gradients to the input data need to be computed. Typically this is not needed. But this might be needed when implementing composition of modules. force_rebind : bool Default is ``False``. This function does nothing if the executors are already bound. But with this ``True``, the executors will be forced to rebind. shared_module : Module Default is ``None``. Currently shared module is not supported for `SequentialModule`. grad_req : str, list of str, dict of str to str Requirement for gradient accumulation. Can be 'write', 'add', or 'null' (default to 'write'). Can be specified globally (str) or for each argument (list, dict). """ if self.binded and not force_rebind: self.logger.warning('Already bound, ignoring bind()') return if inputs_need_grad: assert for_training is True assert shared_module is None, 'Shared module is not supported' assert len(self._modules) > 0, 'Attempting to bind an empty SequentialModule' self.binded = True # the same label shapes are used for all chained modules self._label_shapes = label_shapes my_data_shapes = data_shapes anybody_ever_needs_label = False for i_layer, module in enumerate(self._modules): meta = self._metas[i_layer] if SequentialModule.META_TAKE_LABELS in meta and \ meta[SequentialModule.META_TAKE_LABELS]: my_label_shapes = label_shapes anybody_ever_needs_label = True else: my_label_shapes = None my_inputs_need_grad = bool(inputs_need_grad or (for_training and i_layer > 0)) if meta.get(SequentialModule.META_AUTO_WIRING, False): data_names = module.data_names assert len(data_names) == len(my_data_shapes) my_data_shapes = [(new_name, shape) for (new_name, (_, shape)) in zip(data_names, my_data_shapes)] module.bind(data_shapes=my_data_shapes, label_shapes=my_label_shapes, for_training=for_training, inputs_need_grad=my_inputs_need_grad, force_rebind=force_rebind, shared_module=None, grad_req=grad_req) # the output of the previous module is the data of the next module my_data_shapes = module.output_shapes if not anybody_ever_needs_label: # then I do not need label either self._label_shapes = None
[docs] def init_optimizer(self, kvstore='local', optimizer='sgd', optimizer_params=(('learning_rate', 0.01),), force_init=False): """Installs and initializes optimizers. Parameters ---------- kvstore : str or KVStore Default `'local'`. optimizer : str or Optimizer Default `'sgd'` optimizer_params : dict Default ``(('learning_rate', 0.01),)``. The default value is not a dictionary, just to avoid pylint warning of dangerous default values. force_init : bool Default ``False``, indicating whether we should force re-initializing the optimizer in the case an optimizer is already installed. """ assert self.binded and self.params_initialized if self.optimizer_initialized and not force_init: self.logger.warning('optimizer already initialized, ignoring.') return for module in self._modules: module.init_optimizer(kvstore=kvstore, optimizer=optimizer, optimizer_params=optimizer_params, force_init=force_init) self.optimizer_initialized = True
[docs] def forward(self, data_batch, is_train=None): """Forward computation. Parameters ---------- data_batch : DataBatch is_train : bool Default is ``None``, in which case `is_train` is take as ``self.for_training``. """ assert self.binded and self.params_initialized # make a shallow copy, just to maintain necessary properties (if any) like # bucket_key, pad, etc. data_batch = copy.copy(data_batch) for i_layer, module in enumerate(self._modules): module.forward(data_batch, is_train=is_train) if i_layer+1 == len(self._modules): # the last layer, do not need to do the followings break data_batch.data = module.get_outputs() if hasattr(data_batch, 'provide_data'): # need to update this, in case the internal module is using bucketing # or whatever data_names = [x[0] for x in module.output_shapes] assert len(data_names) == len(data_batch.data) data_batch.provide_data = [(name, x.shape) for name, x in zip(data_names, data_batch.data)]
[docs] def backward(self, out_grads=None): """Backward computation.""" assert self.binded and self.params_initialized for i_layer, module in reversed(list(zip(range(len(self._modules)), self._modules))): module.backward(out_grads=out_grads) if i_layer == 0: break out_grads = module.get_input_grads()
[docs] def update(self): """Updates parameters according to installed optimizer and the gradient computed in the previous forward-backward cycle. """ assert self.binded and self.params_initialized and self.optimizer_initialized for module in self._modules: module.update()
[docs] def get_outputs(self, merge_multi_context=True): """Gets outputs from a previous forward computation. Parameters ---------- merge_multi_context : bool Default is ``True``. In the case when data-parallelism is used, the outputs will be collected from multiple devices. A ``True`` value indicate that we should merge the collected results so that they look like from a single executor. Returns ------- list of NDArray or list of list of NDArray If `merge_multi_context` is ``True``, it is like ``[out1, out2]``. Otherwise, it is like ``[[out1_dev1, out1_dev2], [out2_dev1, out2_dev2]]``. All the output elements are numpy arrays. """ assert self.binded and self.params_initialized return self._modules[-1].get_outputs(merge_multi_context=merge_multi_context)
[docs] def get_input_grads(self, merge_multi_context=True): """Gets the gradients with respect to the inputs of the module. Parameters ---------- merge_multi_context : bool Default is ``True``. In the case when data-parallelism is used, the outputs will be collected from multiple devices. A ``True`` value indicate that we should merge the collected results so that they look like from a single executor. Returns ------- list of NDArrays or list of list of NDArrays If `merge_multi_context` is ``True``, it is like ``[grad1, grad2]``. Otherwise, it is like ``[[grad1_dev1, grad1_dev2], [grad2_dev1, grad2_dev2]]``. All the output elements are `NDArray`. """ assert self.binded and self.params_initialized and self.inputs_need_grad return self._modules[0].get_input_grads(merge_multi_context=merge_multi_context)
[docs] def update_metric(self, eval_metric, labels, pre_sliced=False): """Evaluates and accumulates evaluation metric on outputs of the last forward computation. Parameters ---------- eval_metric : EvalMetric labels : list of NDArray Typically ``data_batch.label``. """ assert self.binded and self.params_initialized for meta, module in zip(self._metas, self._modules): if SequentialModule.META_TAKE_LABELS in meta and \ meta[SequentialModule.META_TAKE_LABELS]: module.update_metric(eval_metric, labels, pre_sliced)
[docs] def install_monitor(self, mon): """Installs monitor on all executors.""" assert self.binded for module in self._modules: module.install_monitor(mon)