| /external/tensorflow/tensorflow/python/ops/ |
| D | nn_xent_test.py | 40 def _SigmoidCrossEntropyWithLogits(self, logits, targets): argument
41 assert len(logits) == len(targets)
42 pred = [1 / (1 + exp(-x)) for x in logits]
52 logits = constant_op.constant(x, shape=sizes, dtype=dtype, name="logits")
55 return logits, targets, losses
60 logits, targets, _ = self._Inputs()
62 labels=targets, logits=logits, name="mylogistic")
69 logits, targets, losses = self._Inputs(dtype=dtype)
71 labels=targets, logits=logits)
80 logits, targets, losses = self._Inputs(dtype=dtype, sizes=[2, 2, 2])
[all …]
|
| D | ctc_ops.py | 82 logits=None): argument
194 logits,
205 logits=None, argument
217 inputs = deprecation.deprecated_argument_lookup("logits", logits, "inputs",
629 def ctc_loss_and_grad(logits, labels, label_length, logit_length, unique=None): argument
653 num_labels = _get_dim(logits, 2)
656 ilabel_log_probs = nn_ops.log_softmax(logits)
679 max_logit_length = _get_dim(logits, 0)
697 def _ctc_loss_op_standard(labels, logits, logit_length, logits_time_major, argument
699 part_before = logits[:, :, :blank_index]
[all …]
|
| D | nn_impl.py | 118 logits=None,
164 labels, logits)
167 with ops.name_scope(name, "logistic_loss", [logits, labels]) as name:
168 logits = ops.convert_to_tensor(logits, name="logits")
171 labels.get_shape().assert_is_compatible_with(logits.get_shape())
174 (logits.get_shape(), labels.get_shape()))
184 zeros = array_ops.zeros_like(logits, dtype=logits.dtype)
185 cond = (logits >= zeros)
186 relu_logits = array_ops.where(cond, logits, zeros)
187 neg_abs_logits = array_ops.where(cond, -logits, logits)
[all …]
|
| /external/tensorflow/tensorflow/python/ops/distributions/ |
| D | bernoulli.py | 52 logits=None, argument
86 logits=logits,
104 def logits(self): member in Bernoulli
139 event = math_ops.cast(event, self.logits.dtype)
140 logits = self.logits
144 def _broadcast(logits, event): argument
145 return (array_ops.ones_like(event) * logits,
146 array_ops.ones_like(logits) * event)
149 logits.get_shape().is_fully_defined() and
150 event.get_shape() == logits.get_shape()):
[all …]
|
| D | categorical.py | 163 logits=None, argument
194 with ops.name_scope(name, values=[logits, probs]) as name:
196 logits=logits,
250 def logits(self): member in Categorical
263 return self.logits.get_shape()[:-1]
272 if self.logits.get_shape().ndims == 2:
273 logits_2d = self.logits
275 logits_2d = array_ops.reshape(self.logits, [-1, self.event_size])
311 k, logits = _broadcast_cat_event_and_params(
312 k, self.logits, base_dtype=self.dtype.base_dtype)
[all …]
|
| D | multinomial.py | 162 logits=None, argument
195 with ops.name_scope(name, values=[total_count, logits, probs]) as name:
202 logits=logits,
225 def logits(self): member in Multinomial
252 self.logits[..., 0], dtype=n_draws.dtype) * n_draws
253 logits = array_ops.ones_like(
254 n_draws[..., array_ops.newaxis], dtype=self.logits.dtype) * self.logits
257 flat_logits = array_ops.reshape(logits, [-1, k]) # [B1B2...Bm, k]
262 logits, n_draw = args[0], args[1] # [K], []
263 x = random_ops.multinomial(logits[array_ops.newaxis, ...], n_draw,
[all …]
|
| /external/tensorflow/tensorflow/examples/speech_commands/ |
| D | models_test.py | 56 logits, dropout_rate = models.create_model(
58 self.assertIsNotNone(logits)
60 self.assertIsNotNone(sess.graph.get_tensor_by_name(logits.name))
68 logits = models.create_model(fingerprint_input, model_settings, "conv",
70 self.assertIsNotNone(logits)
71 self.assertIsNotNone(sess.graph.get_tensor_by_name(logits.name))
78 logits, dropout_rate = models.create_model(
80 self.assertIsNotNone(logits)
82 self.assertIsNotNone(sess.graph.get_tensor_by_name(logits.name))
90 logits, dropout_rate = models.create_model(
[all …]
|
| /external/tensorflow/tensorflow/python/kernel_tests/random/ |
| D | multinomial_op_test.py | 40 def composed_sampler(logits, num_samples): argument
42 unif = random_ops.random_uniform(logits.get_shape().concatenate(
46 logits = array_ops.expand_dims(logits, -1)
49 return math_ops.argmax(logits + noise, axis=1)
63 logits = constant_op.constant([[-10., 10., -10.], [-10., -10., 10.]])
66 logits, num_samples, output_dtype=output_dtype))
102 logits = np.array([[1000.] * 5])
104 logits *= -1
105 samples = self.evaluate(random_ops.multinomial(logits, 10))
121 logits = np.log(probs).astype(np.float32)
[all …]
|
| /external/tensorflow/tensorflow/python/kernel_tests/ |
| D | losses_test.py | 120 logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0],
125 losses.softmax_cross_entropy(labels, logits, weights=None)
130 logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0],
133 loss = losses.softmax_cross_entropy(labels, logits)
139 logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0],
144 loss = losses.softmax_cross_entropy(labels, logits)
150 logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0],
155 loss = losses.softmax_cross_entropy(labels, logits, weights)
160 logits = constant_op.constant([[10.0, 0.0, 0.0], [0.0, 10.0, 0.0],
165 loss = losses.softmax_cross_entropy(labels, logits,
[all …]
|
| D | ctc_loss_op_test.py | 77 logits=inputs,
323 logits = random_ops.random_uniform([num_frames, batch_size, num_labels])
333 t.watch(logits)
336 logits=logits,
339 ref_grad = t.gradient(ref_loss, [logits])
348 grad = gradients_impl.gradients(loss, [logits])
355 logits=logits,
370 logits = random_ops.random_uniform([num_frames, batch_size, num_labels])
385 logits=logits,
388 ctc_loss_grads = gradients_impl.gradients(ctc_loss, [logits])[0]
[all …]
|
| D | sparse_xent_op_test.py | 151 labels=[[0, 2]], logits=[[0., 1.], [2., 3.], [2., 3.]])
157 labels=constant_op.constant(0), logits=constant_op.constant(1.0))
163 labels=labels, logits=[[7.]])
170 labels=constant_op.constant(0), logits=constant_op.constant([1.0]))
208 labels=l, logits=f, name="xent")
238 labels=l, logits=f, name="xent"), [f])[0]
243 labels=l, logits=f, name="xent"), [f])[0]
266 labels=labels, logits=features)
275 labels=labels, logits=features), [features])[0]
297 logits = array_ops.placeholder(dtypes.float32, shape=[None, 3])
[all …]
|
| /external/tensorflow/tensorflow/python/ops/losses/ |
| D | losses_impl.py | 321 def hinge_loss(labels, logits, weights=1.0, scope=None, argument
355 if logits is None:
357 with ops.name_scope(scope, "hinge_loss", (logits, labels, weights)) as scope:
358 logits = math_ops.cast(logits, dtype=dtypes.float32)
360 logits.get_shape().assert_is_compatible_with(labels.get_shape())
365 math_ops.subtract(all_ones, math_ops.multiply(labels, logits)))
659 multi_class_labels, logits, weights=1.0, label_smoothing=0, scope=None, argument
703 if logits is None:
706 (logits, multi_class_labels, weights)) as scope:
707 logits = ops.convert_to_tensor(logits)
[all …]
|
| /external/tensorflow/tensorflow/core/kernels/ |
| D | sparse_xent_op.cc | 57 const Tensor& logits = context->input(0); in Compute() local
59 OP_REQUIRES(context, TensorShapeUtils::IsMatrix(logits.shape()), in Compute()
61 logits.shape().DebugString())); in Compute()
65 OP_REQUIRES(context, logits.dim_size(0) == labels.dim_size(0), in Compute()
69 logits.shape().DebugString(), " and labels shape ", in Compute()
71 OP_REQUIRES(context, logits.dim_size(1) > 0, in Compute()
74 logits.shape().DebugString())); in Compute()
85 {0}, 1, logits.shape(), &back_out)); in Compute()
87 if (logits.dim_size(0) > 0) { in Compute()
90 context, CheckInvalidLabelIndex<Index>(labels, logits.dim_size(1))); in Compute()
[all …]
|
| D | sparse_xent_op_gpu.cu.cc | 42 typename TTypes<T>::ConstMatrix logits, in Compute()
46 const int rows = logits.dimension(kBatchDim); in Compute()
47 const int cols = logits.dimension(kClassDim); in Compute()
53 ctx, maximum.data(), logits.data(), 2, rows, cols, 1, 1, constants.kOne, in Compute()
62 void operator()(OpKernelContext* ctx, typename TTypes<T>::ConstMatrix logits, in operator ()()
66 SparseXentEigenImpl<GPUDevice, T, Index>::Compute(ctx, logits, labels, in operator ()()
|
| D | softmax_op_functor.h | 34 void operator()(const Device& d, typename TTypes<T>::ConstMatrix logits,
44 static void Compute(const Device& d, typename TTypes<T>::ConstMatrix logits, in Compute()
49 const int batch_size = logits.dimension(kBatchDim); in Compute()
50 const int num_classes = logits.dimension(kClassDim); in Compute()
66 auto shifted_logits = (logits - logits.maximum(along_class) in Compute()
|
| D | sparse_xent_op.h | 60 typename TTypes<const T, 2>::Tensor32Bit logits, in SparseXentLossGenerator() argument
64 : logits_(logits), in SparseXentLossGenerator()
139 typename TTypes<T>::ConstMatrix logits, in Compute()
148 To32Bit(maximum).device(d) = To32Bit(logits).maximum(along_row); in Compute()
162 void operator()(OpKernelContext* ctx, typename TTypes<T>::ConstMatrix logits,
174 typename TTypes<T>::ConstMatrix logits, in Compute()
186 const int batch_size = logits.dimension(kBatchDim); in Compute()
187 const int num_classes = logits.dimension(kClassDim); in Compute()
213 RowMaxReduction<Device, T>::Compute(ctx, logits, scratch); in Compute()
218 To32Bit(logits) - in Compute()
|
| /external/tensorflow/tensorflow/python/kernel_tests/distributions/ |
| D | categorical_test.py | 41 logits = random_ops.random_uniform(
43 return categorical.Categorical(logits, dtype=dtype)
54 self.assertAllEqual([2], dist.logits.get_shape())
59 logits = np.log(p) - 50.
60 dist = categorical.Categorical(logits=logits)
63 self.assertAllEqual([2], dist.logits.get_shape())
65 self.assertAllClose(dist.logits, logits)
101 self.assertEqual(dist.logits.dtype, dtypes.float32)
102 self.assertEqual(dist.logits.dtype, dist.entropy().dtype)
104 dist.logits.dtype, dist.prob(np.array(
[all …]
|
| /external/tensorflow/tensorflow/lite/kernels/ |
| D | multinomial.cc | 37 const FloatType* logits, int logits_size, in MultinomialSample() argument
48 max_logit = std::max(max_logit, logits[i]); in MultinomialSample()
52 FloatType odds = std::exp(logits[i] - max_logit) + last_odds; in MultinomialSample()
76 const FloatType* logits, int logits_size, in MultinomialSample() argument
82 rng, logits, logits_size, in MultinomialSample()
87 rng, logits, logits_size, in MultinomialSample()
100 const TfLiteTensor* logits, int logits_offset, in MultinomialSample() argument
103 switch (logits->type) { in MultinomialSample()
112 context, rng, GetTensorData<float>(logits) + logits_offset, in MultinomialSample()
119 context, rng, GetTensorData<double>(logits) + logits_offset, in MultinomialSample()
[all …]
|
| D | multinomial_test.cc | 57 MultinomialOpModel(tflite::TensorData logits, int num_samples, in MultinomialOpModel() argument
59 logits_ = AddInput(logits); in MultinomialOpModel()
70 int logits() { return logits_; } in logits() function in tflite::__anon1d5c03730111::MultinomialOpModel
109 m.PopulateTensor<Float>(m.logits(), in TYPED_TEST()
140 m.PopulateTensor<Float>(m.logits(), in TYPED_TEST()
165 m.logits(), {static_cast<Float>(-1000.0f), static_cast<Float>(-1000.0f), in TYPED_TEST()
188 std::vector<Float> logits(kNumLogits, static_cast<Float>(0.0f)); in TYPED_TEST() local
189 m.PopulateTensor<Float>(m.logits(), logits); in TYPED_TEST()
214 std::vector<Float> logits( in TYPED_TEST() local
216 m.PopulateTensor<Float>(m.logits(), logits); in TYPED_TEST()
|
| /external/tensorflow/tensorflow/python/eager/benchmarks/resnet50/ |
| D | hvp_test.py | 37 logits = model(images, training=True)
39 logits=logits, onehot_labels=labels)
49 logits = model(images, training=True)
51 logits=logits, onehot_labels=labels)
57 logits = model(images, training=True)
59 logits=logits, onehot_labels=labels)
70 logits = model(images, training=True)
72 logits=logits, onehot_labels=labels)
|
| /external/tensorflow/tensorflow/compiler/tf2xla/kernels/ |
| D | softmax_op.cc | 43 xla::XlaOp logits, bool log) { in BuildSoftmaxCustomCall() argument
44 TF_ASSIGN_OR_RETURN(xla::Shape logits_shape, b->GetShape(logits)); in BuildSoftmaxCustomCall()
45 return xla::CustomCallWithLayout(b, log ? "log_softmax" : "softmax", {logits}, in BuildSoftmaxCustomCall()
69 auto logits = ctx->Input(0); in Compile() local
76 b->ReportErrorOrReturn(BuildSoftmaxCustomCall(b, logits, log_)); in Compile()
85 xla::Reduce(logits, xla::MinValue(b, xla_type), max_func, {kClassDim}); in Compile()
88 auto shifted_logits = xla::Sub(logits, logits_max, batch_dims); in Compile()
118 xla::XlaOp logits, xla::XlaOp labels) { in CrossEntropyWithLogits() argument
127 xla::Reduce(logits, xla::MinValue(b, xla_type), max_func, {kClassDim}); in CrossEntropyWithLogits()
131 auto shifted_logits = xla::Sub(logits, logits_max, {kBatchDim}); in CrossEntropyWithLogits()
[all …]
|
| /external/tensorflow/tensorflow/python/tpu/ |
| D | async_checkpoint_test.py | 69 logits = math_ops.matmul(features, w)
70 loss = losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)
83 def metric_fn(labels, logits): argument
85 logging.info('LABELS %s %s', labels, logits)
87 'recall@1': metrics_lib.recall_at_k(labels, logits, 1),
88 'recall@5': metrics_lib.recall_at_k(labels, logits, 5),
91 loss = losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)
92 eval_metrics = (metric_fn, [labels, logits])
|
| /external/tensorflow/tensorflow/core/api_def/base_api/ |
| D | api_def_BoostedTreesPredict.pbtxt | 12 name: "logits"
14 Output rank 2 Tensor containing logits for each example.
26 scalar, dimension of the logits, to be used for partial logits
32 computes the logits. It is designed to be used during prediction.
|
| D | api_def_Softmax.pbtxt | 4 name: "logits"
12 Same shape as `logits`.
19 $$softmax[i, j] = exp(logits[i, j]) / sum_j(exp(logits[i, j]))$$
|
| D | api_def_LogSoftmax.pbtxt | 4 name: "logits"
12 Same shape as `logits`.
19 logsoftmax[i, j] = logits[i, j] - log(sum(exp(logits[i])))
|