text
stringlengths
81
112k
Gets a new grouped symbol whose output contains inputs to output nodes of the original symbol. Example ------- >>> x = mx.sym.Variable('x') >>> y = mx.sym.Variable('y') >>> z = mx.sym.Variable('z') >>> a = y+z >>> b = x+a >>> b.get_children() ...
Lists all the arguments in the symbol. Example ------- >>> a = mx.sym.var('a') >>> b = mx.sym.var('b') >>> c = a + b >>> c.list_arguments ['a', 'b'] Returns ------- args : list of string List containing the names of all the ar...
Lists all the outputs in the symbol. Example ------- >>> a = mx.sym.var('a') >>> b = mx.sym.var('b') >>> c = a + b >>> c.list_outputs() ['_plus12_output'] Returns ------- list of str List of all the outputs. For mo...
Lists all the auxiliary states in the symbol. Example ------- >>> a = mx.sym.var('a') >>> b = mx.sym.var('b') >>> c = a + b >>> c.list_auxiliary_states() [] Example of auxiliary states in `BatchNorm`. >>> data = mx.symbol.Variable('data') ...
Lists all arguments and auxiliary states of this Symbol. Returns ------- inputs : list of str List of all inputs. Examples -------- >>> bn = mx.sym.BatchNorm(name='bn') >>> bn.list_arguments() ['bn_data', 'bn_gamma', 'bn_beta'] >>> bn...
Infers the type of all arguments and all outputs, given the known types for some arguments. This function takes the known types of some arguments in either positional way or keyword argument way as input. It returns a tuple of `None` values if there is not enough information to deduce t...
The actual implementation for calling type inference API. def _infer_type_impl(self, partial, *args, **kwargs): """The actual implementation for calling type inference API.""" # pylint: disable=too-many-locals if len(args) != 0 and len(kwargs) != 0: raise ValueError('Can only specif...
Infers the shapes of all arguments and all outputs given the known shapes of some arguments. This function takes the known shapes of some arguments in either positional way or keyword argument way as input. It returns a tuple of `None` values if there is not enough information to deduce...
The actual implementation for calling shape inference API. def _infer_shape_impl(self, partial, *args, **kwargs): """The actual implementation for calling shape inference API.""" # pylint: disable=too-many-locals if len(args) != 0 and len(kwargs) != 0: raise ValueError('Can only spe...
Saves symbol to a file. You can also use pickle to do the job if you only work on python. The advantage of `load`/`save` functions is that the file contents are language agnostic. This means the model saved by one language binding can be loaded by a different language binding of `MXNet`...
Saves symbol to a JSON string. See Also -------- symbol.load_json : Used to load symbol from JSON string. def tojson(self): """Saves symbol to a JSON string. See Also -------- symbol.load_json : Used to load symbol from JSON string. """ json_str...
Helper function to get NDArray lists handles from various inputs. Parameters ---------- arg_key : str The name of argument, used for error message. args : list of NDArray or dict of str to NDArray Input arguments to the symbols. If type is list of ND...
Bind current symbol to get an executor, allocate all the arguments needed. Allows specifying data types. This function simplifies the binding procedure. You need to specify only input data shapes. Before binding the executor, the function allocates arguments and auxiliary states that we...
Binds the current symbol to an executor and returns it. We first declare the computation and then bind to the data to run. This function returns an executor which provides method `forward()` method for evaluation and a `outputs()` method to get all the results. Example ------- ...
Gets the autodiff of current symbol. This function can only be used if current symbol is a loss function. .. note:: This function is currently not implemented. Parameters ---------- wrt : Array of String keyword arguments of the symbol that the gradients are taken....
Evaluates a symbol given arguments. The `eval` method combines a call to `bind` (which returns an executor) with a call to `forward` (executor method). For the common use case, where you might repeatedly evaluate with same arguments, eval is slow. In that case, you should call `...
Return symbol for target backend. Parameters ---------- backend : str The backend names. Returns ------- out : Symbol The created Symbol for target backend. def get_backend_symbol(self, backend): """Return symbol for target backend. ...
Perform pixel-shuffling on the input. def hybrid_forward(self, F, x): """Perform pixel-shuffling on the input.""" f = self._factor # (N, C*f, W) x = F.reshape(x, (0, -4, -1, f, 0)) # (N, C, f, W) x = F.transpose(x, (0, 1, 3, 2)) # (N, C,...
Perform pixel-shuffling on the input. def hybrid_forward(self, F, x): """Perform pixel-shuffling on the input.""" f1, f2 = self._factors # (N, f1*f2*C, H, W) x = F.reshape(x, (0, -4, -1, f1 * f2, 0, 0)) # (N, C, f1*f2, H, W) x = F.r...
Perform pixel-shuffling on the input. def hybrid_forward(self, F, x): """Perform pixel-shuffling on the input.""" # `transpose` doesn't support 8D, need other implementation f1, f2, f3 = self._factors # (N, C*f1*f2*f3, D, H, W) ...
Retry calling the decorated function using an exponential backoff. http://www.saltycrane.com/blog/2009/11/trying-out-retry-decorator-python/ original from: http://wiki.python.org/moin/PythonDecoratorLibrary#Retry :param target_exception: the exception to check. may be a tuple of exceptions to chec...
Returns a module loaded with the provided model. Parameters ---------- model_name: str Prefix of the MXNet model name as stored on the local directory. epoch_num : int Epoch number of model we would like to load. input_shape: tuple The shape of the input data in the form o...
Creates a new MXNet module. Parameters ---------- sym : Symbol An MXNet symbol. input_shape: tuple The shape of the input data in the form of (batch_size, channels, height, width) files: list of strings List of URLs pertaining to files that need to be downloaded in order t...
evalute network given validation record file Parameters: ---------- net : str or None Network name or use None to load from json without modifying path_imgrec : str path to the record validation file path_imglist : str path to the list file to replace labels in record file, ...
Initializes the parameters and auxiliary states. By default this function does nothing. Subclass should override this method if contains parameters. Parameters ---------- initializer : Initializer Called to initialize parameters if needed. arg_params : dict ...
Evaluates and accumulates evaluation metric on outputs of the last forward computation. Subclass should override this method if needed. Parameters ---------- eval_metric : EvalMetric labels : list of NDArray Typically ``data_batch.label``. def update_metric(self, ev...
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...
Forward computation. Here we do nothing but to keep a reference to the scores and the labels so that we can do backward computation. Parameters ---------- data_batch : DataBatch Could be anything with similar API implemented. is_train : bool Default is ``...
Actual implementation of the backward computation. The computation should take ``self._scores`` and ``self._labels`` and then compute the gradients with respect to the scores, store it as an `NDArray` in ``self._scores_grad``. Instead of defining a subclass and overriding this function,...
Encode sentences and (optionally) build a mapping from string tokens to integer indices. Unknown keys will be added to vocabulary. Parameters ---------- sentences : list of list of str A list of sentences to encode. Each sentence should be a list of string tokens. vocab : None o...
Resets the iterator to the beginning of the data. def reset(self): """Resets the iterator to the beginning of the data.""" self.curr_idx = 0 random.shuffle(self.idx) for buck in self.data: np.random.shuffle(buck) self.nddata = [] self.ndlabel = [] fo...
Returns the next batch of data. def next(self): """Returns the next batch of data.""" if self.curr_idx == len(self.idx): raise StopIteration i, j = self.idx[self.curr_idx] self.curr_idx += 1 if self.major_axis == 1: data = self.nddata[i][j:j+self.batch_s...
Returns the singleton instance. Upon its first call, it creates a new instance of the decorated class and calls its `__init__` method. On all subsequent calls, the already created instance is returned. def getInstance(self): """ Returns the singleton instance. Upon its first call, it cr...
Description : run lipnet training code using argument info def main(): """ Description : run lipnet training code using argument info """ parser = argparse.ArgumentParser() parser.add_argument('--batch_size', type=int, default=64) parser.add_argument('--image_path', type=str, default='./data/da...
Get the variable given a name if one exists or create a new one if missing. Parameters ---------- name : str name of the variable **kwargs : more arguments that's passed to symbol.Variable def get(self, name, **kwargs): """Get the variable given a name i...
Reset before re-using the cell for another graph. def reset(self): """Reset before re-using the cell for another graph.""" self._init_counter = -1 self._counter = -1 if hasattr(self, '_cells'): for cell in self._cells: cell.reset()
Initial state for this cell. Parameters ---------- func : callable, default symbol.zeros Function for creating initial state. Can be symbol.zeros, symbol.uniform, symbol.Variable etc. Use symbol.Variable if you want to directly feed input as state...
Unpack fused weight matrices into separate weight matrices. For example, say you use a module object `mod` to run a network that has an lstm cell. In `mod.get_params()[0]`, the lstm parameters are all represented as a single big vector. `cell.unpack_weights(mod.get_params()[0])` will un...
Pack separate weight matrices into a single packed weight. Parameters ---------- args : dict of str -> NDArray Dictionary containing unpacked weights. Returns ------- args : dict of str -> NDArray Dictionary with packed weights associated...
Unroll an RNN cell across time steps. Parameters ---------- length : int Number of steps to unroll. inputs : Symbol, list of Symbol, or None If `inputs` is a single Symbol (usually the output of Embedding symbol), it should have shape (bat...
Get activation function. Convert if is string def _get_activation(self, inputs, activation, **kwargs): """Get activation function. Convert if is string""" if isinstance(activation, string_types): return symbol.Activation(inputs, act_type=activation, **kwargs) else: retur...
slice fused rnn weights def _slice_weights(self, arr, li, lh): """slice fused rnn weights""" args = {} gate_names = self._gate_names directions = self._directions b = len(directions) p = 0 for layer in range(self._num_layers): for direction in direct...
Unfuse the fused RNN in to a stack of rnn cells. Returns ------- cell : mxnet.rnn.SequentialRNNCell unfused cell that can be used for stepping, and can run on CPU. def unfuse(self): """Unfuse the fused RNN in to a stack of rnn cells. Returns ------- ...
Append a cell into the stack. Parameters ---------- cell : BaseRNNCell The cell to be appended. During unroll, previous cell's output (or raw inputs if no previous cell) is used as the input to this cell. def add(self, cell): """Append a cell into the stack. ...
Reads an image from file path or URL, optionally resizing to given image dimensions and subtracting mean. :param img_path: path to file, or url to download :param image_dims: image dimensions to resize to, or None :param mean: mean file to subtract, or None :return: loaded image, in RGB format def ...
Changes device of given mxnet arguments :param arg_params: arguments :param aux_params: auxiliary parameters :param ctx: new device context :return: arguments and auxiliary parameters on new device def _ch_dev(arg_params, aux_params, ctx): """ Changes device of given mxnet arguments :param ...
Run the layer comparison on a caffe model, given its prototxt, weights and mean. The comparison is done by inferring on a given image using both caffe and mxnet model :param image_url: image file or url to run inference on :param gpu: gpu to use, -1 for cpu :param caffe_prototxt_path: path to caffe prot...
Implementation of Breadth-first search (BFS) on caffe network DAG :param root_node: root node of caffe network DAG :param process_node: function to run on each node def _bfs(root_node, process_node): """ Implementation of Breadth-first search (BFS) on caffe network DAG :param root_node: root node o...
Compare layer by layer of a caffe network with mxnet network :param caffe_net: loaded caffe network :param arg_params: arguments :param aux_params: auxiliary parameters :param exe: mxnet model :param layer_name_to_record: map between caffe layer and information record :param top_to_layers: map b...
Entrypoint for compare_layers def main(): """Entrypoint for compare_layers""" parser = argparse.ArgumentParser( description='Tool for testing caffe to mxnet conversion layer by layer') parser.add_argument('--image_url', type=str, default='https://github.com/dmlc/web-data/ra...
Get executor to Stochastic Gradient Langevin Dynamics and/or Bayesian Dark Knowledge def get_executor(sym, ctx, data_inputs, initializer=None): """Get executor to Stochastic Gradient Langevin Dynamics and/or Bayesian Dark Knowledge""" data_shapes = {k: v.shape for k, v in data_inputs.items()} arg_names = s...
Create copy of parameters def copy_param(exe, new_param=None): """Create copy of parameters""" if new_param is None: new_param = {k: nd.empty(v.shape, ctx=mx.cpu()) for k, v in exe.arg_dict.items()} for k, v in new_param.items(): exe.arg_dict[k].copyto(v) return new_param
Parse command line arguments def parse_args(): """Parse command line arguments""" parser = argparse.ArgumentParser() parser.add_argument("font_path", help="Path to ttf font file or directory containing ttf files") parser.add_argument("--loss", help="'ctc' or 'warpctc' loss [Default 'ctc']", default='ct...
Program entry point def main(): """Program entry point""" args = parse_args() if not any(args.loss == s for s in ['ctc', 'warpctc']): raise ValueError("Invalid loss '{}' (must be 'ctc' or 'warpctc')".format(args.loss)) hp = Hyperparams() # Start a multiprocessor captcha image generator ...
Gatys et al. CVPR 2017 ref: Image Style Transfer Using Convolutional Neural Networks def optimize(args): """ Gatys et al. CVPR 2017 ref: Image Style Transfer Using Convolutional Neural Networks """ if args.cuda: ctx = mx.gpu(0) else: ctx = mx.cpu(0) # load the content and...
Get symbol of mnist def get_mnist_sym(output_op=None, num_hidden=400): """Get symbol of mnist""" net = mx.symbol.Variable('data') net = mx.symbol.FullyConnected(data=net, name='mnist_fc1', num_hidden=num_hidden) net = mx.symbol.Activation(data=net, name='mnist_relu1', act_type="relu") net = mx.symb...
Get synthetic gradient value def synthetic_grad(X, theta, sigma1, sigma2, sigmax, rescale_grad=1.0, grad=None): """Get synthetic gradient value""" if grad is None: grad = nd.empty(theta.shape, theta.context) theta1 = theta.asnumpy()[0] theta2 = theta.asnumpy()[1] v1 = sigma1 ** 2 v2 = s...
Get toy symbol def get_toy_sym(teacher=True, teacher_noise_precision=None): """Get toy symbol""" if teacher: net = mx.symbol.Variable('data') net = mx.symbol.FullyConnected(data=net, name='teacher_fc1', num_hidden=100) net = mx.symbol.Activation(data=net, name='teacher_relu1', act_type=...
Run DistilledSGLD on mnist dataset def run_mnist_DistilledSGLD(num_training=50000, gpu_id=None): """Run DistilledSGLD on mnist dataset""" X, Y, X_test, Y_test = load_mnist(num_training) minibatch_size = 100 if num_training >= 10000: num_hidden = 800 total_iter_num = 1000000 teac...
Run SGLD on toy dataset def run_toy_SGLD(gpu_id=None): """Run SGLD on toy dataset""" X, Y, X_test, Y_test = load_toy() minibatch_size = 1 teacher_noise_precision = 1.0 / 9.0 net = get_toy_sym(True, teacher_noise_precision) data_shape = (minibatch_size,) + X.shape[1::] data_inputs = {'data':...
Run DistilledSGLD on toy dataset def run_toy_DistilledSGLD(gpu_id): """Run DistilledSGLD on toy dataset""" X, Y, X_test, Y_test = load_toy() minibatch_size = 1 teacher_noise_precision = 1.0 teacher_net = get_toy_sym(True, teacher_noise_precision) student_net = get_toy_sym(False) data_shape ...
Run HMC on toy dataset def run_toy_HMC(gpu_id=None): """Run HMC on toy dataset""" X, Y, X_test, Y_test = load_toy() minibatch_size = Y.shape[0] noise_precision = 1 / 9.0 net = get_toy_sym(True, noise_precision) data_shape = (minibatch_size,) + X.shape[1::] data_inputs = {'data': nd.zeros(da...
Run synthetic SGLD def run_synthetic_SGLD(): """Run synthetic SGLD""" theta1 = 0 theta2 = 1 sigma1 = numpy.sqrt(10) sigma2 = 1 sigmax = numpy.sqrt(2) X = load_synthetic(theta1=theta1, theta2=theta2, sigmax=sigmax, num=100) minibatch_size = 1 total_iter_num = 1000000 lr_scheduler...
wrapper function for loading pascal voc dataset Parameters: ---------- image_set : str train, trainval... year : str 2007, 2012 or combinations splitted by comma devkit_path : str root directory of dataset shuffle : bool whether to shuffle initial list Retur...
wrapper function for loading ms coco dataset Parameters: ---------- image_set : str train2014, val2014, valminusminival2014, minival2014 dirname: str root dir for coco shuffle: boolean initial shuffle def load_coco(image_set, dirname, shuffle=False): """ wrapper fun...
Resets the iterator to the beginning of the data. def reset(self): """Resets the iterator to the beginning of the data.""" self.curr_idx = 0 #shuffle data in each bucket random.shuffle(self.idx) for i, buck in enumerate(self.sentences): self.indices[i], self.sentence...
Returns the next batch of data. def next(self): """Returns the next batch of data.""" if self.curr_idx == len(self.idx): raise StopIteration #i = batches index, j = starting record i, j = self.idx[self.curr_idx] self.curr_idx += 1 indices = self.ndindex[i][...
Converts a reshape layer from mxnet to coreml. This doesn't currently handle the deprecated parameters for the reshape layer. Parameters ---------- network: net An mxnet network object. layer: node Node to convert. module: module A module for MXNet builder: Neura...
Convert a transpose layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_transpose(net...
Convert a flatten layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_flatten(net, no...
Convert a softmax layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_softmax(net, no...
Convert an activation layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_activation(...
Convert a leakyrelu layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_leakyrelu(net...
Convert an elementwise add layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_elemen...
Convert a convolution layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_convolution...
Convert a pooling layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_pooling(net, no...
Convert a batchnorm layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_batchnorm(net...
Convert concat layer from mxnet to coreml. Parameters ---------- network: net A mxnet network object. layer: node Node to convert. module: module An module for MXNet builder: NeuralNetworkBuilder A neural network builder object. def convert_concat(net, node, ...
convert from mxnet's opts to dmlc's opts def dmlc_opts(opts): """convert from mxnet's opts to dmlc's opts """ args = ['--num-workers', str(opts.num_workers), '--num-servers', str(opts.num_servers), '--cluster', opts.launcher, '--host-file', opts.hostfile, '--...
Unfuses the fused RNN in to a stack of rnn cells. def _unfuse(self): """Unfuses the fused RNN in to a stack of rnn cells.""" assert not self._projection_size, "_unfuse does not support projection layer yet!" assert not self._lstm_state_clip_min and not self._lstm_state_clip_max, \ ...
Initial state for this cell. Parameters ---------- batch_size: int Only required for `NDArray` API. Size of the batch ('N' in layout). Dimension of the input. func : callable, default `ndarray.zeros` Function for creating initial state. F...
forward using CUDNN or CPU kenrel def _forward_kernel(self, F, inputs, states, **kwargs): """ forward using CUDNN or CPU kenrel""" if self._layout == 'NTC': inputs = F.swapaxes(inputs, dim1=0, dim2=1) if self._projection_size is None: params = (kwargs['{}{}_{}_{}'.format...
Wait for network service to appear @param server: host to connect to (str) @param port: port (int) @param timeout: in seconds, if None or 0 wait forever @return: True of False, if timeout is None may return only True or throw unhandled network exception def wait_ssh_ope...
Wait for network service to appear @param server: host to connect to (str) @param port: port (int) @param timeout: in seconds, if None or 0 wait forever @return: True of False, if timeout is None may return only True or throw unhandled network exception def wait_port_op...
Convert symbol for detail information. Parameters ---------- symbol: Symbol Symbol to be visualized. shape: dict A dict of shapes, str->shape (tuple), given input shapes. line_length: int Rotal length of printed lines positions: list Relative or absolute position...
Creates a visualization (Graphviz digraph object) of the given computation graph. Graphviz must be installed for this function to work. Parameters ---------- title: str, optional Title of the generated visualization. symbol: Symbol A symbol from the computation graph. The generated ...
Measure the accuracy of ResNet Parameters ---------- data_iterator: Iter examples of dataset network: ResNet Returns ---------- tuple of array element def evaluate_accuracy(data_iterator, network): """ Measure the accuracy of ResNet Parameters ---------- data_...
Training with multiple GPUs Parameters ---------- batch_list: List list of dataset context: List a list of all GPUs to be used for training network: ResNet gluon_trainer: rain module of gluon def train_batch(batch_list, context, network, gluon_trainer): """ Training...
Take an executor's underlying symbol graph and return its generated optimized version. Parameters ---------- executor : An executor for which you want to see an optimized symbol. Getting an optimized symbol is useful to compare and verify the work TensorRT has done against a legacy behaviou...
Bind current symbol to get an optimized trt executor. Parameters ---------- symbol : Symbol The symbol you wish to bind, and optimize with TensorRT. ctx : Context The device context the generated executor to run on. all_params : Dict of str->ndarray A dictionary of mapping...
Parameters ---------- num_classes : int, default 1000 Number of classification classes. num_layers : int Number of layers for the variant of densenet. Options are 11, 13, 16, 19. batch_norm : bool, default False Use batch normalization. dtype: str, float32 or float16 ...
:param frame: an (w,h,channels) numpy array (image) :return: DataBatch of (1,channels,data_shape,data_shape) def create_batch(self, frame): """ :param frame: an (w,h,channels) numpy array (image) :return: DataBatch of (1,channels,data_shape,data_shape) """ frame_resize =...
detect all images in iterator Parameters: ---------- det_iter : DetIter iterator for all testing images show_timer : Boolean whether to print out detection exec time Returns: ---------- list of detection results def detect_iter(self, det...
Return detections for batch :param batch: :return: def detect_batch(self, batch): """ Return detections for batch :param batch: :return: """ self.mod.forward(batch, is_train=False) detections = self.mod.get_outputs()[0] positive_detections...
wrapper for detecting multiple images Parameters: ---------- im_list : list of str image path or list of image paths root_dir : str directory of input images, optional if image path already has full directory information extension : str ...
visualize detections in one image Parameters: ---------- img : numpy.array image, in bgr format dets : numpy.array ssd detections, numpy.array([[id, score, x1, y1, x2, y2]...]) each row is one object classes : tuple or list of str ...
First column (class id) is -1 for negative detections :param detections: :return: def filter_positive_detections(detections): """ First column (class id) is -1 for negative detections :param detections: :return: """ class_idx = 0 assert(isinstance...
wrapper for im_detect and visualize_detection Parameters: ---------- im_list : list of str or str image path or list of image paths root_dir : str or None directory of input images, optional if image path already has full directory information ...
Runs the caffe upgrade tool on the prototxt to create a prototxt in the latest format. This enable us to work just with latest structures, instead of supporting all the variants :param caffe_root: link to caffe root folder, where the upgrade tool is located :param deploy_proto: name of the original prototx...
Reads from the caffe prototxt the network structure :param processed_deploy_prototxt: name of prototxt to load, preferably the prototxt should be processed before using a call to process_network_proto() :return: network_def, layer_name_to_record, top_to_layers network_def: caffe network structure, give...