Skip to content

Conversation

@jacquesqiao
Copy link
Member

@jacquesqiao jacquesqiao commented Aug 18, 2017

fix: #3557 use low level API to construct a mnist training logic

  1. fix sgd-op [done] 8b3d33a
  2. add rowwiseadd-grad-op [done] "add rowwise add backward op" #3183 82b820e
  3. add mul-grad-op [done] "mul op gradient" #3379
  4. fix math matmul [done] fix math matmul #3577
  5. fix rowwiseadd-grad-op [done] Fix rowwise add #3604
  6. register rowwise_grad_gpu_op [done] register rowwise add gpu kernel #3660

@jacquesqiao jacquesqiao changed the title init minst init mnist Aug 18, 2017
return cost_name


def get_backward_net(forward_net):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

create_backward_net

return net


def print_inputs_outputs(op):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

debug_print_op

var = scope.new_var(param_name)
tensor = var.get_tensor()
tensor.set_dims(dims)
data = numpy.random.uniform(
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

as a demo, maybe we better use the built-in operator?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

dzhwinter
dzhwinter previously approved these changes Aug 24, 2017
Copy link
Contributor

@dzhwinter dzhwinter left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM++

cost_grad.set(numpy.ones(cost_shape).astype("float32"), place)


def mean_cost(cost):
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

mean cost get the cross_entropy output for debugging. we'd better rename this snippet.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

@jacquesqiao jacquesqiao merged commit b9a4b2e into PaddlePaddle:develop Aug 24, 2017
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

write mnist training with the current python API

3 participants