Python Model.get_top_bricks方法代码示例

# 需要导入模块: from blocks.model import Model [as 别名]
# 或者: from blocks.model.Model import get_top_bricks [as 别名]
def main():

    import configurations
    from stream import DStream
    logger = logging.getLogger(__name__)
    cfig = getattr(configurations, 'get_config_penn')()

    rnnlm = Rnnlm(cfig['vocabsize'], cfig['nemb'], cfig['nhids'])
    rnnlm.weights_init = IsotropicGaussian(0.1)
    rnnlm.biases_init = Constant(0.)
    rnnlm.push_initialization_config()
    rnnlm.generator.transition.weights_init = Orthogonal()

    sentence = tensor.lmatrix('sentence')
    sentence_mask = tensor.matrix('sentence_mask')
    batch_cost = rnnlm.cost(sentence, sentence_mask).sum()
    batch_size = sentence.shape[1].copy(name='batch_size')
    cost = aggregation.mean(batch_cost, batch_size)
    cost.name = "sequence_log_likelihood"
    logger.info("Cost graph is built")

    model = Model(cost)
    parameters = model.get_parameter_dict()
    logger.info("Parameters:\n" +
                pprint.pformat(
                    [(key, value.get_value().shape) for key, value
                        in parameters.items()],
                    width=120))

    for brick in model.get_top_bricks():
        brick.initialize()
    cg = ComputationGraph(cost)
    algorithm = GradientDescent(
        cost=cost, parameters=cg.parameters,
        step_rule=CompositeRule([StepClipping(10.0), Scale(0.01)]))

    gradient_norm = aggregation.mean(algorithm.total_gradient_norm)
    step_norm = aggregation.mean(algorithm.total_step_norm)
    monitored_vars = [cost, gradient_norm, step_norm]

    train_monitor = TrainingDataMonitoring(variables=monitored_vars, after_batch=True,
                                           before_first_epoch=True, prefix='tra')

    extensions = [train_monitor, Timing(), Printing(after_batch=True),
                  FinishAfter(after_n_epochs=1000),
                  Printing(every_n_batches=1)]

    train_stream = DStream(datatype='train', config=cfig)
    main_loop = MainLoop(model=model,
                         data_stream=train_stream,
                         algorithm=algorithm,
                         extensions=extensions)

    main_loop.run()

# 需要导入模块: from blocks.model import Model [as 别名]
# 或者: from blocks.model.Model import get_top_bricks [as 别名]
def test_model():
    x = tensor.matrix('x')
    mlp1 = MLP([Tanh(), Tanh()], [10, 20, 30], name="mlp1")
    mlp2 = MLP([Tanh()], [30, 40], name="mlp2")
    h1 = mlp1.apply(x)
    h2 = mlp2.apply(h1)

    model = Model(h2)
    assert model.get_top_bricks() == [mlp1, mlp2]
    # The order of parameters returned is deterministic but
    # not sensible.
    assert list(model.get_parameter_dict().items()) == [
        ('/mlp2/linear_0.b', mlp2.linear_transformations[0].b),
        ('/mlp1/linear_1.b', mlp1.linear_transformations[1].b),
        ('/mlp1/linear_0.b', mlp1.linear_transformations[0].b),
        ('/mlp1/linear_0.W', mlp1.linear_transformations[0].W),
        ('/mlp1/linear_1.W', mlp1.linear_transformations[1].W),
        ('/mlp2/linear_0.W', mlp2.linear_transformations[0].W)]

    # Test getting and setting parameter values
    mlp3 = MLP([Tanh()], [10, 10])
    mlp3.allocate()
    model3 = Model(mlp3.apply(x))
    parameter_values = {
        '/mlp/linear_0.W': 2 * numpy.ones((10, 10),
                                          dtype=theano.config.floatX),
        '/mlp/linear_0.b': 3 * numpy.ones(10, dtype=theano.config.floatX)}
    model3.set_parameter_values(parameter_values)
    assert numpy.all(
        mlp3.linear_transformations[0].parameters[0].get_value() == 2)
    assert numpy.all(
        mlp3.linear_transformations[0].parameters[1].get_value() == 3)
    got_parameter_values = model3.get_parameter_values()
    assert len(got_parameter_values) == len(parameter_values)
    for name, value in parameter_values.items():
        assert_allclose(value, got_parameter_values[name])

    # Test exception is raised if parameter shapes don't match
    def helper():
        parameter_values = {
            '/mlp/linear_0.W': 2 * numpy.ones((11, 11),
                                              dtype=theano.config.floatX),
            '/mlp/linear_0.b': 3 * numpy.ones(11, dtype=theano.config.floatX)}
        model3.set_parameter_values(parameter_values)
    assert_raises(ValueError, helper)

    # Test name conflict handling
    mlp4 = MLP([Tanh()], [10, 10])

    def helper():
        Model(mlp4.apply(mlp3.apply(x)))
    assert_raises(ValueError, helper)

# 需要导入模块: from blocks.model import Model [as 别名]
# 或者: from blocks.model.Model import get_top_bricks [as 别名]
def main(mode, save_path, num_batches, data_path=None):
    reverser = WordReverser(100, len(char2code), name="reverser")

    if mode == "train":
        # Data processing pipeline
        dataset_options = dict(dictionary=char2code, level="character",
                               preprocess=_lower)
        if data_path:
            dataset = TextFile(data_path, **dataset_options)
        else:
            dataset = OneBillionWord("training", [99], **dataset_options)
        data_stream = dataset.get_example_stream()
        data_stream = Filter(data_stream, _filter_long)
        data_stream = Mapping(data_stream, reverse_words,
                              add_sources=("targets",))
        data_stream = Batch(data_stream, iteration_scheme=ConstantScheme(10))
        data_stream = Padding(data_stream)
        data_stream = Mapping(data_stream, _transpose)

        # Initialization settings
        reverser.weights_init = IsotropicGaussian(0.1)
        reverser.biases_init = Constant(0.0)
        reverser.push_initialization_config()
        reverser.encoder.weights_init = Orthogonal()
        reverser.generator.transition.weights_init = Orthogonal()

        # Build the cost computation graph
        chars = tensor.lmatrix("features")
        chars_mask = tensor.matrix("features_mask")
        targets = tensor.lmatrix("targets")
        targets_mask = tensor.matrix("targets_mask")
        batch_cost = reverser.cost(
            chars, chars_mask, targets, targets_mask).sum()
        batch_size = chars.shape[1].copy(name="batch_size")
        cost = aggregation.mean(batch_cost, batch_size)
        cost.name = "sequence_log_likelihood"
        logger.info("Cost graph is built")

        # Give an idea of what's going on
        model = Model(cost)
        parameters = model.get_parameter_dict()
        logger.info("Parameters:\n" +
                    pprint.pformat(
                        [(key, value.get_value().shape) for key, value
                         in parameters.items()],
                        width=120))

        # Initialize parameters
        for brick in model.get_top_bricks():
            brick.initialize()

        # Define the training algorithm.
        cg = ComputationGraph(cost)
        algorithm = GradientDescent(
            cost=cost, parameters=cg.parameters,
            step_rule=CompositeRule([StepClipping(10.0), Scale(0.01)]))

        # Fetch variables useful for debugging
        generator = reverser.generator
        (energies,) = VariableFilter(
            applications=[generator.readout.readout],
            name_regex="output")(cg.variables)
        (activations,) = VariableFilter(
            applications=[generator.transition.apply],
            name=generator.transition.apply.states[0])(cg.variables)
        max_length = chars.shape[0].copy(name="max_length")
        cost_per_character = aggregation.mean(
            batch_cost, batch_size * max_length).copy(
                name="character_log_likelihood")
        min_energy = energies.min().copy(name="min_energy")
        max_energy = energies.max().copy(name="max_energy")
        mean_activation = abs(activations).mean().copy(
                name="mean_activation")
        observables = [
            cost, min_energy, max_energy, mean_activation,
            batch_size, max_length, cost_per_character,
            algorithm.total_step_norm, algorithm.total_gradient_norm]
        for name, parameter in parameters.items():
            observables.append(parameter.norm(2).copy(name + "_norm"))
            observables.append(algorithm.gradients[parameter].norm(2).copy(
                name + "_grad_norm"))

        # Construct the main loop and start training!
        average_monitoring = TrainingDataMonitoring(
            observables, prefix="average", every_n_batches=10)
        main_loop = MainLoop(
            model=model,
            data_stream=data_stream,
            algorithm=algorithm,
            extensions=[
                Timing(),
                TrainingDataMonitoring(observables, after_batch=True),
                average_monitoring,
                FinishAfter(after_n_batches=num_batches)
                # This shows a way to handle NaN emerging during
                # training: simply finish it.
                .add_condition(["after_batch"], _is_nan),
                # Saving the model and the log separately is convenient,
                # because loading the whole pickle takes quite some time.
                Checkpoint(save_path, every_n_batches=500,
#.........这里部分代码省略.........

# 需要导入模块: from blocks.model import Model [as 别名]
# 或者: from blocks.model.Model import get_top_bricks [as 别名]

#.........这里部分代码省略.........

    # Build the cost computation graph [steps, batch_size, 3]
    x = T.tensor3('features', dtype=floatX)
    if debug:
        x.tag.test_value = np.ones((max_length,batch_size,3)).astype(floatX)
    x = x[:max_length,:,:]  # has to be after setting test_value
    cost = generator.cost(x)
    cost.name = "sequence_log_likelihood"

    # Give an idea of what's going on
    model = Model(cost)
    params = model.get_params()
    logger.info("Parameters:\n" +
                pprint.pformat(
                    [(key, value.get_value().shape) for key, value
                     in params.items()],
                    width=120))
    model_size = 0
    for v in params.itervalues():
        s = v.get_value().shape
        model_size += s[0] * (s[1] if len(s) > 1 else 1)
    logger.info("Total number of parameters %d"%model_size)

    #------------------------------------------------------------
    extensions = []
    if old_model_name == 'continue':
        extensions.append(LoadFromDump(jobname))
    elif old_model_name:
        # or you can just load the weights without state using:
        old_params = LoadFromDump(old_model_name).manager.load_parameters()
        model.set_param_values(old_params)
    else:
        # Initialize parameters
        for brick in model.get_top_bricks():
            brick.initialize()

    if sample:
        assert old_model_name and old_model_name != 'continue'
        Sample(generator, steps=max_length, path=old_model_name).do(None)
        exit(0)

    #------------------------------------------------------------
    # Define the training algorithm.
    cg = ComputationGraph(cost)
    if dropout > 0.:
        from blocks.roles import INPUT, OUTPUT
        dropout_target = VariableFilter(roles=[OUTPUT],
                                        bricks=transitions,
                                        name_regex='states')(cg.variables)
        print('# dropout %d' % len(dropout_target))
        cg = apply_dropout(cg, dropout_target, dropout)
        opt_cost = cg.outputs[0]
    else:
        opt_cost = cost

    if step_method == 'adam':
        step_rule = Adam(learning_rate)
    elif step_method == 'rmsprop':
        step_rule = RMSProp(learning_rate, decay_rate=0.95)
    elif step_method == 'adagrad':
        step_rule = AdaGrad(learning_rate)
    elif step_method == 'adadelta':
        step_rule = AdaDelta()
    elif step_method == 'scale':
        step_rule = Scale(learning_rate)
    else:

# 需要导入模块: from blocks.model import Model [as 别名]
# 或者: from blocks.model.Model import get_top_bricks [as 别名]
        vocabs = pkl.load(f)
        word_vocab, rel_vocab = vocabs['word'], vocabs['rel']

    with open('dataset/trainXY.json') as f:
        train = json.load(f)
        train = wrap_stream(train)

    with open('dataset/testXY.json') as f:
        test = json.load(f)
        test = wrap_stream(test)

    model = LSTMModel(len(vocabs['word']), n_mem, len(vocabs['rel']))
    cg = ComputationGraph(model.cost)

    bricks_model = Model(model.cost)
    for brick in bricks_model.get_top_bricks():
        brick.initialize()
    model.lookup.W.set_value(vocabs['word'].get_embeddings().astype(theano.config.floatX))

    if dropout:
        pass
        # logger.info('Applying dropout of {}'.format(dropout))
        # lstm_dropout = [v for v in cg.intermediary_variables if v.name in {'W_cell_to_in', 'W_cell_to_out'}]
        # cg = apply_dropout(cg, lstm_dropout, drop_prob=dropout)

    # summary of what's going on
    parameters = bricks_model.get_parameter_dict()
    logger.info("Parameters:\n" +
                pprint.pformat(
                    [(key, value.get_value().shape, value.get_value().mean()) for key, value
                     in parameters.items()],

# 需要导入模块: from blocks.model import Model [as 别名]
# 或者: from blocks.model.Model import get_top_bricks [as 别名]
from blocks.bricks import Tanh, Softmax, Linear, MLP, Identity, Rectifier
from blocks.bricks.lookup import LookupTable

import data
from paramsaveload import SaveLoadParams


config = importlib.import_module('.deepmind_attentive_reader', 'config')
path = os.path.join(os.getenv("DATAPATH"), "deepmind-qa/cnn/questions/training")
vocab_path = os.path.join(os.getenv("DATAPATH"), "deepmind-qa/cnn/stats/training/vocab.txt")
ds, stream = data.setup_datastream(path, vocab_path, config)
model_path = "~/code/deepmind_qa/deepmind_attentive_reader_epoch2step33900.pkl"
m = config.Model(config, ds.vocab_size)
model = Model(m.sgd_cost)
SaveLoadParams(path=model_path, model=model).do_load()
bricks = model.get_top_bricks()
print "brick load completed..."

def get_prediction_function():
	question = tensor.imatrix('question')
	question_mask = tensor.imatrix('question_mask')
	context = tensor.imatrix('context')
	context_mask = tensor.imatrix('context_mask')
	answer = tensor.ivector('answer')
	candidates = tensor.imatrix('candidates')
	candidates_mask = tensor.imatrix('candidates_mask')

	"""	
	question = question.dimshuffle(1, 0)
	question_mask = question_mask.dimshuffle(1, 0)
	context = context.dimshuffle(1, 0)

# 需要导入模块: from blocks.model import Model [as 别名]
# 或者: from blocks.model.Model import get_top_bricks [as 别名]
def main(mode, save_path, num_batches, data_path=None):
    reverser = WordReverser(100, len(char2code), name="reverser")

    if mode == "train":
        # Data processing pipeline
        dataset_options = dict(dictionary=char2code, level="character",
                               preprocess=_lower)
        if data_path:
            dataset = TextFile(data_path, **dataset_options)
        else:
            dataset = OneBillionWord("training", [99], **dataset_options)
        data_stream = dataset.get_example_stream()
        data_stream = Filter(data_stream, _filter_long)
        data_stream = Mapping(data_stream, reverse_words,
                              add_sources=("targets",))
        data_stream = Batch(data_stream, iteration_scheme=ConstantScheme(10))
        data_stream = Padding(data_stream)
        data_stream = Mapping(data_stream, _transpose)

        # Initialization settings
        reverser.weights_init = IsotropicGaussian(0.1)
        reverser.biases_init = Constant(0.0)
        reverser.push_initialization_config()
        reverser.encoder.weights_init = Orthogonal()
        reverser.generator.transition.weights_init = Orthogonal()

        # Build the cost computation graph
        chars = tensor.lmatrix("features")
        chars_mask = tensor.matrix("features_mask")
        targets = tensor.lmatrix("targets")
        targets_mask = tensor.matrix("targets_mask")
        batch_cost = reverser.cost(
            chars, chars_mask, targets, targets_mask).sum()
        batch_size = chars.shape[1].copy(name="batch_size")
        cost = aggregation.mean(batch_cost, batch_size)
        cost.name = "sequence_log_likelihood"
        logger.info("Cost graph is built")

        # Give an idea of what's going on
        model = Model(cost)
        parameters = model.get_parameter_dict()
        logger.info("Parameters:\n" +
                    pprint.pformat(
                        [(key, value.get_value().shape) for key, value
                         in parameters.items()],
                        width=120))

        # Initialize parameters
        for brick in model.get_top_bricks():
            brick.initialize()

        # Define the training algorithm.
        cg = ComputationGraph(cost)
        algorithm = GradientDescent(
            cost=cost, parameters=cg.parameters,
            step_rule=CompositeRule([StepClipping(10.0), Scale(0.01)]))

        # Fetch variables useful for debugging
        generator = reverser.generator
        (energies,) = VariableFilter(
            applications=[generator.readout.readout],
            name_regex="output")(cg.variables)
        (activations,) = VariableFilter(
            applications=[generator.transition.apply],
            name=generator.transition.apply.states[0])(cg.variables)
        max_length = chars.shape[0].copy(name="max_length")
        cost_per_character = 
            aggregation.mean(batch_cost, batch_size * max_length).copy(
            name="character_log_likelihood")
        min_energy = energies.min().copy(name="min_energy")
        max_energy = energies.max().copy(name="max_energy")
        mean_activation = abs(activations).mean() .copy(name="mean_activation")
        observables = [
            cost, min_energy, max_energy, mean_activation,
            batch_size, max_length, cost_per_character,
            algorithm.total_step_norm, algorithm.total_gradient_norm]
        for name, parameter in parameters.items():
            observables.append( parameter.norm(2)
                    .copy(name=name + "_norm"))
            observables.append( algorithm.gradients[parameter].norm(2)
                    .copy(name=name + "_grad_norm"))

        # Construct the main loop and start training!
        average_monitoring = TrainingDataMonitoring(
            observables, prefix="average", every_n_batches=10)
        main_loop = MainLoop(
            model=model,
            data_stream=data_stream,
            algorithm=algorithm,
            extensions=[
                Timing(),
                TrainingDataMonitoring(observables, after_batch=True),
                average_monitoring,
                FinishAfter(after_n_batches=num_batches)
                # This shows a way to handle NaN emerging during
                # training: simply finish it.
                .add_condition(["after_batch"], _is_nan),
                # Saving the model and the log separately is convenient,
                # because loading the whole pickle takes quite some time.
                Checkpoint(save_path, every_n_batches=500,
#.........这里部分代码省略.........