我正在尝试使用tensorflow_probability构建一个 mcmc 链。这是我的代码：
chain_states, kernel_results = tfp.mcmc.sample_chain(
num_results=tf.constant(1e3, dtype=tf.int32),
num_burnin_steps=tf.constant(1e2, dtype=tf.int32),
parallel_iterations=tf.constant(10, dtype=tf.int32),
current_state=current_state,
kernel=tfp.mcmc.MetropolisHastings(
inner_kernel=tfp.mcmc.HamiltonianMonteCarlo(
target_log_prob_fn=joint_log_prob,
num_leapfrog_steps=tf.constant(2, dtype=tf.int32),
step_size=tf.Variable(1.),
step_size_update_fn=tfp.mcmc.make_simple_step_size_update_policy()
)))
但我得到了这个错误::
> InvalidArgumentError Traceback (most recent call last) <ipython-input-13-7e972cc65053> in <module>()
> ----> 1 make_model(well_complex, well_ligand, fi_complex, fi_ligand)
>
> ~/Documents/GitHub/assaytools2/assaytools2/assaytools2/inference.py in
> make_model(well_complex, well_ligand, fi_complex, fi_ligand)
> 162 num_leapfrog_steps=tf.constant(2, dtype=tf.int32),
> 163 step_size=tf.Variable(1.),
> --> 164 step_size_update_fn=tfp.mcmc.make_simple_step_size_update_policy()
> 165 )))
> 166
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow_probability/python/mcmc/sample.py
> in sample_chain(num_results, current_state, previous_kernel_results,
> kernel, num_burnin_steps, num_steps_between_results,
> parallel_iterations, name)
> 238
> 239 if previous_kernel_results is None:
> --> 240 previous_kernel_results = kernel.bootstrap_results(current_state)
> 241 return tf.scan(
> 242 fn=_scan_body,
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow_probability/python/mcmc/metropolis_hastings.py
> in bootstrap_results(self, init_state)
> 261 name=mcmc_util.make_name(self.name, 'mh', 'bootstrap_results'),
> 262 values=[init_state]):
> --> 263 pkr = self.inner_kernel.bootstrap_results(init_state)
> 264 if not has_target_log_prob(pkr):
> 265 raise ValueError(
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow_probability/python/mcmc/hmc.py
> in bootstrap_results(self, init_state)
> 506 def bootstrap_results(self, init_state):
> 507 """Creates initial `previous_kernel_results` using a supplied `state`."""
> --> 508 kernel_results = self._impl.bootstrap_results(init_state)
> 509 if self.step_size_update_fn is not None:
> 510 step_size_assign = self.step_size_update_fn(self.step_size, None) # pylint:
> disable=not-callable
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow_probability/python/mcmc/metropolis_hastings.py
> in bootstrap_results(self, init_state)
> 261 name=mcmc_util.make_name(self.name, 'mh', 'bootstrap_results'),
> 262 values=[init_state]):
> --> 263 pkr = self.inner_kernel.bootstrap_results(init_state)
> 264 if not has_target_log_prob(pkr):
> 265 raise ValueError(
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow_probability/python/mcmc/hmc.py
> in bootstrap_results(self, init_state)
> 672 init_target_log_prob,
> 673 init_grads_target_log_prob,
> --> 674 ] = mcmc_util.maybe_call_fn_and_grads(self.target_log_prob_fn, init_state)
> 675 return UncalibratedHamiltonianMonteCarloKernelResults(
> 676 log_acceptance_correction=tf.zeros_like(init_target_log_prob),
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow_probability/python/mcmc/util.py
> in maybe_call_fn_and_grads(fn, fn_arg_list, result, grads,
> check_non_none_grads, name)
> 232 fn_arg_list = (list(fn_arg_list) if is_list_like(fn_arg_list)
> 233 else [fn_arg_list])
> --> 234 result, grads = _value_and_gradients(fn, fn_arg_list, result, grads)
> 235 if not all(r.dtype.is_floating
> 236 for r in (result if is_list_like(result) else [result])): # pylint: disable=superfluous-parens
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow_probability/python/mcmc/util.py
> in _value_and_gradients(fn, fn_arg_list, result, grads, name)
> 207 ]
> 208 else:
> --> 209 grads = tfe.gradients_function(fn)(*fn_arg_list)
> 210 else:
> 211 if is_list_like(result) and len(result) == len(fn_arg_list):
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/eager/backprop.py
> in decorated(*args, **kwds)
> 368 """Computes the gradient of the decorated function."""
> 369
> --> 370 _, grad = val_and_grad_function(f, params=params)(*args, **kwds)
> 371 return grad
> 372
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/eager/backprop.py
> in decorated(*args, **kwds)
> 469 "receive keyword arguments.")
> 470 val, vjp = make_vjp(f, params)(*args, **kwds)
> --> 471 return val, vjp(dy=dy)
> 472
> 473 return decorated
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/eager/backprop.py
> in vjp(dy)
> 539 return imperative_grad.imperative_grad(
> 540 _default_vspace, this_tape, nest.flatten(result), sources,
> --> 541 output_gradients=dy)
> 542 return result, vjp
> 543
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/eager/imperative_grad.py
> in imperative_grad(vspace, tape, target, sources, output_gradients)
> 61 """
> 62 return pywrap_tensorflow.TFE_Py_TapeGradient(
> ---> 63 tape._tape, vspace, target, sources, output_gradients) # pylint: disable=protected-access
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/eager/backprop.py
> in _gradient_function(op_name, attr_tuple, num_inputs, inputs,
> outputs, out_grads)
> 115 return [None] * num_inputs
> 116
> --> 117 return grad_fn(mock_op, *out_grads)
> 118
> 119
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/ops/math_grad.py
> in _ProdGrad(op, grad)
> 158 with ops.device("/cpu:0"):
> 159 rank = array_ops.rank(op.inputs[0])
> --> 160 reduction_indices = (reduction_indices + rank) % rank
> 161 reduced = math_ops.cast(reduction_indices, dtypes.int32)
> 162 idx = math_ops.range(0, rank)
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py
> in binary_op_wrapper(x, y)
> 860 with ops.name_scope(None, op_name, [x, y]) as name:
> 861 if isinstance(x, ops.Tensor) and isinstance(y, ops.Tensor):
> --> 862 return func(x, y, name=name)
> 863 elif not isinstance(y, sparse_tensor.SparseTensor):
> 864 try:
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/tensorflow/python/ops/gen_math_ops.py
> in add(x, y, name)
> 322 else:
> 323 message = e.message
> --> 324 _six.raise_from(_core._status_to_exception(e.code, message), None)
> 325
> 326
>
> ~/anaconda2/envs/py36/lib/python3.6/site-packages/six.py in
> raise_from(value, from_value)
>
> InvalidArgumentError: cannot compute Add as input #0(zero-based) was
> expected to be a int32 tensor but is a int64 tensor [Op:Add] name:
> mcmc_sample_chain/mh_bootstrap_results/mh_bootstrap_results/hmc_kernel_bootstrap_results/maybe_call_fn_and_grads/value_and_gradients/add/
I doubled checked and none of my initial tensors were of integer type.
I wonder where I did it wrong.
Thanks!