Python dynet.ParameterCollection方法代碼示例

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def add_params(model, size, name=""):
    """ Adds parameters to the model.

    Inputs:
        model (dy.ParameterCollection): The parameter collection for the model.
        size (tuple of int): The size to create.
        name (str, optional): The name of the parameters.
    """
    if len(size) == 1:
        print("vector " + name + ": " +
              str(size[0]) + "; uniform in [-0.1, 0.1]")
    else:
        print("matrix " +
              name +
              ": " +
              str(size[0]) +
              " x " +
              str(size[1]) +
              "; uniform in [-0.1, 0.1]")
    return model.add_parameters(size,
                                init=dy.UniformInitializer(0.1),
                                name=name) 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def __init__(self, vocab, options):
        import dynet as dy
        from uuparser.feature_extractor import FeatureExtractor
        global dy
        self.model = dy.ParameterCollection()
        self.trainer = dy.AdamTrainer(self.model, alpha=options.learning_rate)
        self.activations = {'tanh': dy.tanh, 'sigmoid': dy.logistic, 'relu':
                            dy.rectify, 'tanh3': (lambda x:
                                                  dy.tanh(dy.cwise_multiply(dy.cwise_multiply(x, x), x)))}
        self.activation = self.activations[options.activation]
        self.costaugFlag = options.costaugFlag
        self.feature_extractor = FeatureExtractor(self.model, options, vocab)
        self.labelsFlag=options.labelsFlag
        mlp_in_dims = options.lstm_output_size*2

        self.unlabeled_MLP = biMLP(self.model, mlp_in_dims, options.mlp_hidden_dims,
                                 options.mlp_hidden2_dims, 1, self.activation)
        if self.labelsFlag:
            self.labeled_MLP = biMLP(self.model, mlp_in_dims, options.mlp_hidden_dims,
                               options.mlp_hidden2_dims,len(self.feature_extractor.irels),self.activation)

        self.proj = options.proj 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def __init__(self):
        self.params = dy.ParameterCollection() 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def renew_collection(self):
        """Renew every time new architecture is sampled to clear out old parameters"""
        self.params = dy.ParameterCollection() 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def init_params(self):
        self.pc = dy.ParameterCollection() 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def create_multilayer_lstm_params(num_layers,
                                  in_size,
                                  state_size,
                                  model,
                                  name=""):
    """ Adds a multilayer LSTM to the model parameters.

    Inputs:
        num_layers (int): Number of layers to create.
        in_size (int): The input size to the first layer.
        state_size (int): The size of the states.
        model (dy.ParameterCollection): The parameter collection for the model.
        name (str, optional): The name of the multilayer LSTM.
    """
    params = []
    in_size = in_size
    state_size = state_size
    for i in range(num_layers):
        layer_name = name + "-" + str(i)
        print(
            "LSTM " +
            layer_name +
            ": " +
            str(in_size) +
            " x " +
            str(state_size) +
            "; default Dynet initialization of hidden weights")
        params.append(dy.VanillaLSTMBuilder(1,
                                            in_size,
                                            state_size,
                                            model))
        in_size = state_size

    return params 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def construct_token_predictor(parameter_collection,
                              params,
                              vocabulary,
                              attention_key_size,
                              snippet_size,
                              anonymizer=None):
    """ Constructs a token predictor given the parameters.

    Inputs:
        parameter_collection (dy.ParameterCollection): Contains the parameters.
        params (dictionary): Contains the command line parameters/hyperparameters.
        vocabulary (Vocabulary): Vocabulary object for output generation.
        attention_key_size (int): The size of the attention keys.
        anonymizer (Anonymizer): An anonymization object.
    """
    if params.use_snippets and anonymizer and not params.previous_decoder_snippet_encoding:
        return SnippetAnonymizationTokenPredictor(parameter_collection,
                                                  params,
                                                  vocabulary,
                                                  attention_key_size,
                                                  snippet_size,
                                                  anonymizer)
    elif params.use_snippets and not params.previous_decoder_snippet_encoding:
        return SnippetTokenPredictor(parameter_collection,
                                     params,
                                     vocabulary,
                                     attention_key_size,
                                     snippet_size)
    elif anonymizer:
        return AnonymizationTokenPredictor(parameter_collection,
                                           params,
                                           vocabulary,
                                           attention_key_size,
                                           anonymizer)
    else:
        return TokenPredictor(parameter_collection,
                              params,
                              vocabulary,
                              attention_key_size) 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def __init__(self, vocab, options):

        # import here so we don't load Dynet if just running parser.py --help for example
        from uuparser.multilayer_perceptron import MLP
        from uuparser.feature_extractor import FeatureExtractor
        import dynet as dy
        global dy

        global LEFT_ARC, RIGHT_ARC, SHIFT, SWAP
        LEFT_ARC, RIGHT_ARC, SHIFT, SWAP = 0,1,2,3

        self.model = dy.ParameterCollection()
        self.trainer = dy.AdamTrainer(self.model, alpha=options.learning_rate)

        self.activations = {'tanh': dy.tanh, 'sigmoid': dy.logistic, 'relu':
                            dy.rectify, 'tanh3': (lambda x:
                            dy.tanh(dy.cwise_multiply(dy.cwise_multiply(x, x), x)))}
        self.activation = self.activations[options.activation]

        self.oracle = options.oracle

        self.headFlag = options.headFlag
        self.rlMostFlag = options.rlMostFlag
        self.rlFlag = options.rlFlag
        self.k = options.k

        #dimensions depending on extended features
        self.nnvecs = (1 if self.headFlag else 0) + (2 if self.rlFlag or self.rlMostFlag else 0)
        self.feature_extractor = FeatureExtractor(self.model, options, vocab, self.nnvecs)
        self.irels = self.feature_extractor.irels

        if options.no_bilstms > 0:
            mlp_in_dims = options.lstm_output_size*2*self.nnvecs*(self.k+1)
        else:
            mlp_in_dims = self.feature_extractor.lstm_input_size*self.nnvecs*(self.k+1)

        self.unlabeled_MLP = MLP(self.model, 'unlabeled', mlp_in_dims, options.mlp_hidden_dims,
                                 options.mlp_hidden2_dims, 4, self.activation)
        self.labeled_MLP = MLP(self.model, 'labeled' ,mlp_in_dims, options.mlp_hidden_dims,
                               options.mlp_hidden2_dims,2*len(self.irels)+2,self.activation) 

# 需要導入模塊: import dynet [as 別名]
# 或者: from dynet import ParameterCollection [as 別名]
def __init__(self):
        super().__init__()

        self.model = dy.ParameterCollection()

        input_size = FEATURES

        # Create word embeddings and initialise
        self.id_to_token = []
        self.token_to_id = {}
        pretrained = []
        if args.word_vectors:
            for line in open(args.word_vectors):
                parts = line.strip().split()
                word = parts[0].lower()
                vector = [float(v) for v in parts[1:]]
                self.token_to_id[word] = len(self.id_to_token)
                self.id_to_token.append(word)
                pretrained.append(vector)
            NWORDS = len(self.id_to_token)
            DIM_WORDS = len(pretrained[0])
            self.pEmbedding = self.model.add_lookup_parameters((NWORDS, DIM_WORDS))
            self.pEmbedding.init_from_array(np.array(pretrained))
            input_size += 4 * DIM_WORDS

        self.hidden = []
        self.bias = []
        self.hidden.append(self.model.add_parameters((HIDDEN, input_size)))
        self.bias.append(self.model.add_parameters((HIDDEN,)))
        for i in range(args.layers - 1):
            self.hidden.append(self.model.add_parameters((HIDDEN, HIDDEN)))
            self.bias.append(self.model.add_parameters((HIDDEN,)))
        self.final_sum = self.model.add_parameters((HIDDEN, 1))