Python tensor.bmatrix方法代码示例

# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import bmatrix [as 别名]
def test_all_grad(self):
        x = tensor.bmatrix('x')
        x_all = x.all()
        gx = theano.grad(x_all, x)
        f = theano.function([x], gx)
        x_random = self.rng.binomial(n=1, p=0.5, size=(5, 7)).astype('int8')
        for x_val in (x_random,
                      numpy.zeros_like(x_random),
                      numpy.ones_like(x_random)):
            gx_val = f(x_val)
            assert gx_val.shape == x_val.shape
            assert numpy.all(gx_val == 0) 

# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import bmatrix [as 别名]
def test_any_grad(self):
        x = tensor.bmatrix('x')
        x_all = x.any()
        gx = theano.grad(x_all, x)
        f = theano.function([x], gx)
        x_random = self.rng.binomial(n=1, p=0.5, size=(5, 7)).astype('int8')
        for x_val in (x_random,
                      numpy.zeros_like(x_random),
                      numpy.ones_like(x_random)):
            gx_val = f(x_val)
            assert gx_val.shape == x_val.shape
            assert numpy.all(gx_val == 0) 

# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import bmatrix [as 别名]
def make_node(self, state, time):
        state = T.as_tensor_variable(state)
        time = T.as_tensor_variable(time)
        return theano.Apply(self, [state, time], [T.bmatrix()])
    
    # Python implementation: 

# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import bmatrix [as 别名]
def BuildModel(modelSpecs, forTrain=True):
        rng = np.random.RandomState()

        ## x is for sequential features and y for matrix (or pairwise) features
        x = T.tensor3('x')
        y = T.tensor4('y')

        ## mask for x and y, respectively
        xmask = T.bmatrix('xmask')
        ymask = T.btensor3('ymask')

        xem = None
        ##if any( k in modelSpecs['seq2matrixMode'] for k in ('SeqOnly', 'Seq+SS') ):
        if config.EmbeddingUsed(modelSpecs):
                xem = T.tensor3('xem')
                distancePredictor = ResNet4DistMatrix( rng, seqInput=x, matrixInput=y, mask_seq=xmask, mask_matrix=ymask, embedInput=xem, modelSpecs=modelSpecs )
        else:
                distancePredictor = ResNet4DistMatrix( rng, seqInput=x, matrixInput=y, mask_seq=xmask, mask_matrix=ymask, modelSpecs=modelSpecs )

        ## labelList is a list of label tensors, each having shape (batchSize, seqLen, seqLen) or (batchSize, seqLen, seqLen, valueDims[response] )
        labelList = []
        if forTrain:
                ## when this model is used for training. We need to define the label variable
		for response in modelSpecs['responses']:
			labelType = Response2LabelType(response)
			rValDims = config.responseValueDims[labelType]

			if labelType.startswith('Discrete'):
				if rValDims > 1:
					## if one response is a vector, then we use a 4-d tensor
					## wtensor is for 16bit integer
					labelList.append( T.wtensor4('Tlabel4' + response ) )
				else:
					labelList.append( T.wtensor3('Tlabel4' + response ) )
			else:
				if rValDims > 1:
					labelList.append( T.tensor4('Tlabel4' + response ) )
				else:
					labelList.append( T.tensor3('Tlabel4' + response ) )

        ## weightList is a list of label weight tensors, each having shape (batchSize, seqLen, seqLen)
        weightList = []
        if len(labelList)>0 and modelSpecs['UseSampleWeight']:
                weightList = [ T.tensor3('Tweight4'+response) for response in modelSpecs['responses'] ]

	## for prediction, both labelList and weightList are empty
        return distancePredictor, x, y, xmask, ymask, xem, labelList, weightList 

# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import bmatrix [as 别名]
def __init__(self, nin, nout, nhid, numpy_rng, scale=1.0):
        self.nin = nin
        self.nout = nout
        self.nhid = nhid
        self.numpy_rng = numpy_rng
        self.theano_rng = RandomStreams(1)
        self.scale = np.float32(scale)

        self.inputs = T.fmatrix('inputs')
        self.targets = T.imatrix('targets')
        self.masks = T.bmatrix('masks')
        self.batchsize = self.inputs.shape[0]

        self.inputs_frames = self.inputs.reshape((
            self.batchsize, self.inputs.shape[1]/nin, nin)).dimshuffle(1,0,2)
        self.targets_frames = self.targets.T
        self.masks_frames = self.masks.T

        self.win = theano.shared(value=self.numpy_rng.normal(
            loc=0, scale=0.001, size=(nin, nhid)
        ).astype(theano.config.floatX), name='win')
        self.wrnn = theano.shared(value=self.scale * np.eye(
            nhid, dtype=theano.config.floatX), name='wrnn')
        self.wout = theano.shared(value=self.numpy_rng.uniform(
            low=-0.01, high=0.01, size=(nhid, nout)
        ).astype(theano.config.floatX), name='wout')
        self.bout = theano.shared(value=np.zeros(
            nout, dtype=theano.config.floatX), name='bout')

        self.params = [self.win, self.wrnn, self.wout, self.bout]

        (self.hiddens, self.outputs), self.updates = theano.scan(
            fn=self.step, sequences=self.inputs_frames,
            outputs_info=[self.theano_rng.uniform(low=0, high=1, size=(
                self.batchsize, nhid), dtype=theano.config.floatX), None])

        self.probabilities = T.nnet.softmax(self.outputs.reshape((
            self.outputs.shape[0] * self.outputs.shape[1],
            self.nout)))
        self.probabilities = T.clip(self.probabilities, 1e-6, 1-1e-6)

        self._stepcosts = T.nnet.categorical_crossentropy(
            self.probabilities, self.targets_frames.flatten()).reshape(
                self.targets_frames.shape)

        self._cost = T.switch(T.gt(self.masks_frames, 0), self._stepcosts, 0).mean()
        self._grads = T.grad(self._cost, self.params)

        self.get_classifications = theano.function(
            [self.inputs], T.argmax(self.probabilities.reshape(self.outputs.shape), axis=2).T) 

# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import bmatrix [as 别名]
def __init__(self, K, vocab_size, num_chars, W_init, 
            nhidden, embed_dim, dropout, train_emb, char_dim, use_feat, gating_fn, 
            save_attn=False):
        self.nhidden = nhidden
        self.embed_dim = embed_dim
        self.dropout = dropout
        self.train_emb = train_emb
        self.char_dim = char_dim
        self.learning_rate = LEARNING_RATE
        self.num_chars = num_chars
        self.use_feat = use_feat
        self.save_attn = save_attn
        self.gating_fn = gating_fn

        self.use_chars = self.char_dim!=0
        if W_init is None: W_init = lasagne.init.GlorotNormal().sample((vocab_size, self.embed_dim))

        doc_var, query_var, cand_var = T.itensor3('doc'), T.itensor3('quer'), \
                T.wtensor3('cand')
        docmask_var, qmask_var, candmask_var = T.bmatrix('doc_mask'), T.bmatrix('q_mask'), \
                T.bmatrix('c_mask')
        target_var = T.ivector('ans')
        feat_var = T.imatrix('feat')
        doc_toks, qry_toks= T.imatrix('dchars'), T.imatrix('qchars')
        tok_var, tok_mask = T.imatrix('tok'), T.bmatrix('tok_mask')
        cloze_var = T.ivector('cloze')
        self.inps = [doc_var, doc_toks, query_var, qry_toks, cand_var, target_var, docmask_var,
                qmask_var, tok_var, tok_mask, candmask_var, feat_var, cloze_var]

        self.predicted_probs, predicted_probs_val, self.network, W_emb, attentions = (
                self.build_network(K, vocab_size, W_init))

        self.loss_fn = T.nnet.categorical_crossentropy(self.predicted_probs, target_var).mean()
        self.eval_fn = lasagne.objectives.categorical_accuracy(self.predicted_probs, 
                target_var).mean()

        loss_fn_val = T.nnet.categorical_crossentropy(predicted_probs_val, target_var).mean()
        eval_fn_val = lasagne.objectives.categorical_accuracy(predicted_probs_val, 
                target_var).mean()

        self.params = L.get_all_params(self.network, trainable=True)
        
        updates = lasagne.updates.adam(self.loss_fn, self.params, learning_rate=self.learning_rate)

        self.train_fn = theano.function(self.inps,
                [self.loss_fn, self.eval_fn, self.predicted_probs], 
                updates=updates,
                on_unused_input='warn')
        self.validate_fn = theano.function(self.inps, 
                [loss_fn_val, eval_fn_val, predicted_probs_val]+attentions,
                on_unused_input='warn') 

# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import bmatrix [as 别名]
def __init__(self, nin, nout, nhid, numpy_rng, scale=1.0):
        self.nin = nin
        self.nout = nout
        self.nhid = nhid
        self.numpy_rng = numpy_rng
        self.scale = np.float32(scale)

        self.inputs = T.fmatrix('inputs')
        self.inputs.tag.test_value = numpy_rng.uniform(
            low=-1., high=1.,
            size=(16, 5 * self.nin)
        ).astype(np.float32)
        self.targets = T.fmatrix('targets')
        self.targets.tag.test_value = np.ones(
            (16, 5 * nout), dtype=np.float32)
        self.masks = T.bmatrix('masks')
        self.masks.tag.test_value = np.ones(
            (16, 5), dtype=np.int8)
        self.batchsize = self.inputs.shape[0]

        self.inputs_frames = self.inputs.reshape((
            self.batchsize, self.inputs.shape[1] / nin,
            nin)).dimshuffle(1, 0, 2)
        self.targets_frames = self.targets.reshape((
            self.batchsize, self.targets.shape[1] / nout,
            nout)).dimshuffle(1, 0, 2)
        self.masks_frames = self.masks.T

        self.h0 = theano.shared(value=np.ones(
            nhid, dtype=theano.config.floatX) * np.float32(.5), name='h0')
        self.win = theano.shared(value=self.numpy_rng.normal(
            loc=0, scale=0.001, size=(nin, nhid)
        ).astype(theano.config.floatX), name='win')
        self.wrnn = theano.shared(value=self.scale * np.eye(
            nhid, dtype=theano.config.floatX), name='wrnn')
        self.wout = theano.shared(value=self.numpy_rng.uniform(
            low=-0.01, high=0.01, size=(nhid, nout)
        ).astype(theano.config.floatX), name='wout')
        self.bout = theano.shared(value=np.zeros(
            nout, dtype=theano.config.floatX), name='bout')

        self.params = [self.win, self.wrnn, self.wout, self.bout]

        (self.hiddens, self.outputs), self.updates = theano.scan(
            fn=self.step, sequences=self.inputs_frames,
            outputs_info=[T.alloc(
                self.h0, self.batchsize, self.nhid), None])

        self._stepcosts = T.sum((self.targets_frames - self.outputs)**2, axis=2)
        self._cost = T.switch(self.masks_frames > 0, self._stepcosts, 0).mean()
        self._grads = T.grad(self._cost, self.params)

        self.getoutputs = theano.function(
            [self.inputs], self.outputs)