ASSERT_EQ(num_elem_numerical, num_grad);

  float* dnumerical = new float[num_elem_numerical]{0};
  memcpy(&dnumerical[0], TF_TensorData(numerical_tensor),
         TF_TensorByteSize(numerical_tensor));

  for (int j = 0; j < num_grad; j++) {
    ASSERT_NEAR(dnumerical[j], expected_grad[j], abs_error);
  }
  delete[] dnumerical;
  TF_DeleteTensor(numerical_tensor);
}

{* ../../docs_src/path_params_numeric_validations/tutorial003_an_py310.py hl[10] *}

## Validações numéricas: maior que ou igual { #number-validations-greater-than-or-equal }

Com `Query` e `Path` (e outras que você verá depois) você pode declarar restrições numéricas.

Aqui, com `ge=1`, `item_id` precisará ser um número inteiro “`g`reater than or `e`qual” a `1`.

{* ../../docs_src/path_params_numeric_validations/tutorial003_an_py310.py hl[10] *}

## Validaciones numéricas: mayor o igual { #number-validations-greater-than-or-equal }

Con `Query` y `Path` (y otros que verás más adelante) puedes declarar restricciones numéricas.

Aquí, con `ge=1`, `item_id` necesitará ser un número entero "`g`reater than or `e`qual" a `1`.

#include <memory>

#include "absl/types/span.h"
#include "tensorflow/c/eager/abstract_tensor_handle.h"
#include "tensorflow/c/eager/unified_api_testutil.h"

namespace tensorflow {
namespace gradients {

/* Returns numerical grad inside `dtheta_approx` given `forward` model and
 * parameter specified by `input_index`.
 *
 * I.e. if y = <output of the forward model> and w = inputs[input_index],
 * this will calculate dy/dw numerically.
 *

  vector<float> theta_data(num_elems);
  memcpy(theta_data.data(), TF_TensorData(theta_tensor),
         TF_TensorByteSize(theta_tensor));

  // Initialize space for the numerical gradient.
  vector<float> dtheta_approx(num_elems);

  // Get theta shape and store in theta_dims.
  int num_dims = TF_NumDims(theta_tensor);
  vector<int64_t> theta_dims(num_dims);

   *
   * <p>This is guaranteed to return zero if the dataset contains a single pair of finite values. It
   * is not guaranteed to return zero when the dataset consists of the same pair of values multiple
   * times, due to numerical errors.
   *
   * <h3>Non-finite values</h3>
   *
   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link

   *
   * <p>This is guaranteed to return zero if the dataset contains a single pair of finite values. It
   * is not guaranteed to return zero when the dataset consists of the same pair of values multiple
   * times, due to numerical errors.
   *
   * <h3>Non-finite values</h3>
   *
   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link

   *
   * <p>This is guaranteed to return zero if the dataset contains only exactly one finite value. It
   * is not guaranteed to return zero when the dataset consists of the same value multiple times,
   * due to numerical errors. However, it is guaranteed never to return a negative result.
   *
   * <h3>Non-finite values</h3>
   *
   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link

   *
   * <p>This is guaranteed to return zero if the dataset contains only exactly one finite value. It
   * is not guaranteed to return zero when the dataset consists of the same value multiple times,
   * due to numerical errors. However, it is guaranteed never to return a negative result.
   *
   * <h3>Non-finite values</h3>
   *
   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link

   * itself. In all other cases, the inverse is a transformation such that applying both the
   * original transformation and its inverse to a value gives you the original value give-or-take
   * numerical errors. Calling this method multiple times on the same instance will always return
   * the same instance. Calling this method on the result of calling this method on an instance will
   * always return that original instance.