Error while trying to run "compute_influences" for TorchLogisticRegression

[ntheol]

Hello,

I tried to run sythetic dataset notebook and in the cell where the influences are computed I get this error:
{
"name": "ValueError",
"message": "Using a target size (torch.Size([100])) that is different to the input size (torch.Size([100, 1])) is deprecated. Please ensure they have the same size.",
"stack": "---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
C:\Users\NIKOLA~1\AppData\Local\Temp/ipykernel_9640/2616014207.py in
4 )
5
----> 6 influence_values = compute_influences(
7 differentiable_model=TorchTwiceDifferentiable(model, F.binary_cross_entropy),
8 training_data=train_data_loader,

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\general.py in compute_influences(differentiable_model, training_data, test_data, input_data, inversion_method, influence_type, hessian_regularization, progress, **kwargs)
307 test_data = deepcopy(training_data)
308
--> 309 influence_factors, _ = compute_influence_factors(
310 differentiable_model,
311 training_data,

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\general.py in compute_influence_factors(model, training_data, test_data, inversion_method, hessian_perturbation, progress, **kwargs)
117 rhs = cat(list(test_grads()))
118
--> 119 return solve_hvp(
120 inversion_method,
121 model,

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\inversion.py in solve_hvp(inversion_method, model, training_data, b, hessian_perturbation, **kwargs)
65 """
66
---> 67 return InversionRegistry.call(
68 inversion_method,
69 model,

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\inversion.py in call(cls, inversion_method, model, training_data, b, hessian_perturbation, **kwargs)
201 """
202
--> 203 return cls.get(type(model), inversion_method)(
204 model, training_data, b, hessian_perturbation, **kwargs
205 )

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\inversion.py in wrapper(*args, **kwargs)
154 @functools.wraps(func)
155 def wrapper(*args, **kwargs):
--> 156 return func(*args, **kwargs)
157
158 cls.registry[key] = wrapper

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\torch\torch_differentiable.py in solve_linear(model, training_data, b, hessian_perturbation)
523 all_x.append(x)
524 all_y.append(y)
--> 525 hessian = model.hessian(torch.cat(all_x), torch.cat(all_y))
526 matrix = hessian + hessian_perturbation * torch.eye(
527 model.num_params, device=model.device

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\torch\torch_differentiable.py in hessian(self, x, y)
157 p.detach() for p in self.model.parameters() if p.requires_grad
158 )
--> 159 return torch.func.hessian(model_func)(params)
160
161 @staticmethod

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\eager_transforms.py in wrapper_fn(*args)
1126 return jvp_out
1127
-> 1128 results = vmap(push_jvp, randomness=randomness)(basis)
1129 if has_aux:
1130 results, aux = results

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\vmap.py in wrapped(*args, **kwargs)
432
433 # If chunk_size is not specified.
--> 434 return _flat_vmap(
435 func, batch_size, flat_in_dims, flat_args, args_spec, out_dims, randomness, **kwargs
436 )

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\vmap.py in fn(*args, **kwargs)
37 def fn(*args, **kwargs):
38 with torch.autograd.graph.disable_saved_tensors_hooks(message):
---> 39 return f(*args, **kwargs)
40 return fn
41

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\vmap.py in _flat_vmap(func, batch_size, flat_in_dims, flat_args, args_spec, out_dims, randomness, **kwargs)
617 try:
618 batched_inputs = _create_batched_inputs(flat_in_dims, flat_args, vmap_level, args_spec)
--> 619 batched_outputs = func(*batched_inputs, **kwargs)
620 return _unwrap_batched(batched_outputs, out_dims, vmap_level, batch_size, func)
621 finally:

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\eager_transforms.py in push_jvp(basis)
1117
1118 def push_jvp(basis):
-> 1119 output = _jvp_with_argnums(func, args, basis, argnums=argnums, has_aux=has_aux)
1120 # output[0] is the output of func(*args)
1121 error_if_complex("jacfwd", output[0], is_input=False)

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\vmap.py in fn(*args, **kwargs)
37 def fn(*args, **kwargs):
38 with torch.autograd.graph.disable_saved_tensors_hooks(message):
---> 39 return f(*args, **kwargs)
40 return fn
41

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\eager_transforms.py in _jvp_with_argnums(func, primals, tangents, argnums, strict, has_aux)
963 primals = _wrap_all_tensors(primals, level)
964 duals = _replace_args(primals, duals, argnums)
--> 965 result_duals = func(*duals)
966 if has_aux:
967 if not (isinstance(result_duals, tuple) and len(result_duals) == 2):

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\eager_transforms.py in wrapper_fn(*args)
487 def wrapper_fn(*args):
488 error_if_complex("jacrev", args, is_input=True)
--> 489 vjp_out = _vjp_with_argnums(func, *args, argnums=argnums, has_aux=has_aux)
490 if has_aux:
491 output, vjp_fn, aux = vjp_out

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\vmap.py in fn(*args, **kwargs)
37 def fn(*args, **kwargs):
38 with torch.autograd.graph.disable_saved_tensors_hooks(message):
---> 39 return f(*args, **kwargs)
40 return fn
41

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\_functorch\eager_transforms.py in _vjp_with_argnums(func, argnums, has_aux, *primals)
289 diff_primals = slice_argnums(primals, argnums, as_tuple=False)
290 tree_map(partial(_create_differentiable, level=level), diff_primals)
--> 291 primals_out = func(*primals)
292
293 if has_aux:

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\torch\torch_differentiable.py in model_func(param)
152 strict=True,
153 )
--> 154 return self.loss(outputs, y.to(self.device))
155
156 params = flatten_tensors_to_vector(

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch
n\functional.py in binary_cross_entropy(input, target, weight, size_average, reduce, reduction)
3087 reduction_enum = _Reduction.get_enum(reduction)
3088 if target.size() != input.size():
-> 3089 raise ValueError(
3090 "Using a target size ({}) that is different to the input size ({}) is deprecated. "
3091 "Please ensure they have the same size.".format(target.size(), input.size())

ValueError: Using a target size (torch.Size([100])) that is different to the input size (torch.Size([100, 1])) is deprecated. Please ensure they have the same size."
}
I am pretty sure that I run the notebook exactly as it is.
Let me know If you need any additional infromation

Thank you!

[Xuzzo]

Hello, thanks for the report. Which version of torch are you using?

[ntheol]

Hello, I have torch 2.0.1

[AnesBenmerzoug]

Thanks @ntheol for reporting this issue. We are currently updating and fixing the Influence Functions notebooks so it's highly appreciated. Could you please switch to the fix/435-broken-imports branch and try running the notebook again?

@schroedk Do you perhaps know what could have caused this error?

[ntheol]

@AnesBenmerzoug I run it without any problem from fix/435-broken-imports branch.

[AnesBenmerzoug]

@ntheol That's great! We will hopefully release a new patch version next week. Until then, you should be able to try things out on that branch.

[ntheol]

Thanks!

[ntheol]

Hello again! I tried to run a notebook of my own with a different dataset and I got the same error again unfortunately in the fix branch.
Also right now I started getting the error about the TorchTwiceDifferentialble that I didn't have before.

[Xuzzo]

Could you paste the trace for both errors?

[ntheol]

---

ImportError Traceback (most recent call last)
C:\Users\NIKOLA~1\AppData\Local\Temp/ipykernel_7504/193257936.py in
8 import torch.nn.functional as F
9 import matplotlib.pyplot as plt
---> 10 from pydvl.influence import compute_influences, TorchTwiceDifferentiable
11 from support.shapley import (
12 synthetic_classification_dataset,

ImportError: cannot import name 'TorchTwiceDifferentiable' from 'pydvl.influence' (c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence_init_.py)

[ntheol]

---

ValueError Traceback (most recent call last)
C:\Users\NIKOLA~1\AppData\Local\Temp/ipykernel_4032/3598293409.py in
----> 1 train_influences = compute_influences(
2 TorchTwiceDifferentiable(nn_model, F.binary_cross_entropy),
3 training_data=training_data_loader,
4 test_data=test_data_loader,
5 influence_type="up",

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\general.py in compute_influences(differentiable_model, training_data, test_data, input_data, inversion_method, influence_type, hessian_regularization, progress, **kwargs)
307 test_data = deepcopy(training_data)
308
--> 309 influence_factors, _ = compute_influence_factors(
310 differentiable_model,
311 training_data,

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\general.py in compute_influence_factors(model, training_data, test_data, inversion_method, hessian_perturbation, progress, **kwargs)
117 rhs = cat(list(test_grads()))
118
--> 119 return solve_hvp(
120 inversion_method,
121 model,

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\inversion.py in solve_hvp(inversion_method, model, training_data, b, hessian_perturbation, **kwargs)
65 """
66
---> 67 return InversionRegistry.call(
68 inversion_method,
69 model,

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\inversion.py in call(cls, inversion_method, model, training_data, b, hessian_perturbation, **kwargs)
201 """
202
--> 203 return cls.get(type(model), inversion_method)(
204 model, training_data, b, hessian_perturbation, **kwargs
205 )

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\inversion.py in wrapper(*args, **kwargs)
154 @functools.wraps(func)
155 def wrapper(*args, **kwargs):
--> 156 return func(*args, **kwargs)
157
158 cls.registry[key] = wrapper

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\torch\torch_differentiable.py in solve_linear(model, training_data, b, hessian_perturbation)
523 all_x.append(x)
524 all_y.append(y)
--> 525 hessian = model.hessian(torch.cat(all_x), torch.cat(all_y))
526 matrix = hessian + hessian_perturbation * torch.eye(
527 model.num_params, device=model.device

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\torch\torch_differentiable.py in hessian(self, x, y)
157 p.detach() for p in self.model.parameters() if p.requires_grad
158 )
--> 159 return torch.func.hessian(model_func)(params)
160
161 @staticmethod

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\eager_transforms.py in wrapper_fn(*args)
1126 return jvp_out
1127
-> 1128 results = vmap(push_jvp, randomness=randomness)(basis)
1129 if has_aux:
1130 results, aux = results

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\vmap.py in wrapped(*args, **kwargs)
432
433 # If chunk_size is not specified.
--> 434 return _flat_vmap(
435 func, batch_size, flat_in_dims, flat_args, args_spec, out_dims, randomness, **kwargs
436 )

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\vmap.py in fn(*args, **kwargs)
37 def fn(*args, **kwargs):
38 with torch.autograd.graph.disable_saved_tensors_hooks(message):
---> 39 return f(*args, **kwargs)
40 return fn
41

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\vmap.py in _flat_vmap(func, batch_size, flat_in_dims, flat_args, args_spec, out_dims, randomness, **kwargs)
617 try:
618 batched_inputs = _create_batched_inputs(flat_in_dims, flat_args, vmap_level, args_spec)
--> 619 batched_outputs = func(*batched_inputs, **kwargs)
620 return _unwrap_batched(batched_outputs, out_dims, vmap_level, batch_size, func)
621 finally:

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\eager_transforms.py in push_jvp(basis)
1117
1118 def push_jvp(basis):
-> 1119 output = _jvp_with_argnums(func, args, basis, argnums=argnums, has_aux=has_aux)
1120 # output[0] is the output of func(*args)
1121 error_if_complex("jacfwd", output[0], is_input=False)

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\vmap.py in fn(*args, **kwargs)
37 def fn(*args, **kwargs):
38 with torch.autograd.graph.disable_saved_tensors_hooks(message):
---> 39 return f(*args, **kwargs)
40 return fn
41

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\eager_transforms.py in _jvp_with_argnums(func, primals, tangents, argnums, strict, has_aux)
963 primals = _wrap_all_tensors(primals, level)
964 duals = _replace_args(primals, duals, argnums)
--> 965 result_duals = func(*duals)
966 if has_aux:
967 if not (isinstance(result_duals, tuple) and len(result_duals) == 2):

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\eager_transforms.py in wrapper_fn(*args)
487 def wrapper_fn(*args):
488 error_if_complex("jacrev", args, is_input=True)
--> 489 vjp_out = _vjp_with_argnums(func, *args, argnums=argnums, has_aux=has_aux)
490 if has_aux:
491 output, vjp_fn, aux = vjp_out

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\vmap.py in fn(*args, **kwargs)
37 def fn(*args, **kwargs):
38 with torch.autograd.graph.disable_saved_tensors_hooks(message):
---> 39 return f(*args, **kwargs)
40 return fn
41

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch_functorch\eager_transforms.py in _vjp_with_argnums(func, argnums, has_aux, *primals)
289 diff_primals = slice_argnums(primals, argnums, as_tuple=False)
290 tree_map(partial(_create_differentiable, level=level), diff_primals)
--> 291 primals_out = func(*primals)
292
293 if has_aux:

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\pydvl\influence\torch\torch_differentiable.py in model_func(param)
152 strict=True,
153 )
--> 154 return self.loss(outputs, y.to(self.device))
155
156 params = flatten_tensors_to_vector(

c:\Users\Nikolas Theol\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\functional.py in binary_cross_entropy(input, target, weight, size_average, reduce, reduction)
3087 reduction_enum = _Reduction.get_enum(reduction)
3088 if target.size() != input.size():
-> 3089 raise ValueError(
3090 "Using a target size ({}) that is different to the input size ({}) is deprecated. "
3091 "Please ensure they have the same size.".format(target.size(), input.size())

ValueError: Using a target size (torch.Size([684646])) that is different to the input size (torch.Size([684646, 1])) is deprecated. Please ensure they have the same size.

[Xuzzo]

It looks like you are not in the correct commit. Could you check again that you pulled the latest version of fix/435-broken-imports?

[ntheol]

Yeah you were right, my fault.
Thanks!

[schroedk]

@ntheol the issue is due to using the binary cross entropy function from torch. It matters, if your tensors have shape (n,) or (n, 1). Here a minimal example:

import torch

y_test = torch.rand((5,))
y_pred = torch.rand((5, 1))

# This raises the observed error
torch.nn.functional.binary_cross_entropy(y_test, y_pred)

The DataLoader definition in the notebook did not account for this. It is fixed in #436.