Dev/add rbm network #35

windy-pig · 2025-06-12T12:17:46Z

Description

添加rbm网络

Checklist:

I have read the CONTRIBUTING.md.

Copilot

Pull Request Overview

Adds a Restricted Boltzmann Machine (RBM) implementation for modeling wavefunction magnitudes and a WaveFunctionNormal wrapper that combines the RBM (for magnitude) with an MLP (for phase) and provides sampling utilities.

Introduce RBM class with forward probability estimation and sample (Gibbs sampling).
Define WaveFunctionNormal to compute complex amplitudes via RBM and MLP, plus generate_conf for sample generation.
Utilize pack_int/unpack_int for integer-based configuration encoding.

Comments suppressed due to low confidence (4)

qmb/rbm.py:88

The parameter name mpl_hidden_size appears to be a typo; it should match the attribute mlp_hidden_size for consistency.

            mpl_hidden_size: tuple[int, ...],

qmb/rbm.py:48

There are currently no unit tests for RBM.sample; consider adding tests to verify that Gibbs sampling converges to the expected distribution over several iterations.

    def sample(self, v: torch.Tensor, k: int = 1) -> torch.Tensor:

qmb/rbm.py:42

[nitpick] The variable name mid2 is not descriptive; consider renaming it to something like hidden_pre_activation or hidden_linear_output for clarity.

        mid2 = torch.nn.functional.linear(v, self.weights.T, self.hidden_bias)  # pylint: disable=E1102

qmb/rbm.py:65

[nitpick] The intermediate name midh is ambiguous; renaming it to hidden_pre_activation would improve readability.

            midh = torch.nn.functional.linear(v, self.weights.T, self.hidden_bias)  # pylint: disable=E1102

Copilot · 2025-06-16T11:29:09Z

qmb/rbm.py

+        self.visible_dim = visible_dim
+        self.hidden_dim = hidden_dim
+        self.weights = torch.nn.Parameter(torch.Tensor(self.visible_dim, self.hidden_dim))
+        init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim).cuda())


Avoid hard-coding .cuda() to keep the code device-agnostic; instead, derive the device from the model parameters (e.g., torch.tensor(self.visible_dim, device=self.weights.device)).

Suggested change

init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim).cuda())

init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim, device=self.weights.device))

你不应该在这里.cuda()，但这个copilot说的也不对，你应该直接

init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim)）

因为模型移动到不同的device（如gpu），是在model创建之后进行的。

Copilot · 2025-06-16T11:29:09Z

qmb/rbm.py

+    ) -> None:
+        super().__init__()
+        self.sites: int = sites
+        assert physical_dim == 2  # ??


[nitpick] This bare assert lacks a message and the trailing # ?? is unclear; consider raising a ValueError with a descriptive message or adding an assertion message.

Suggested change

assert physical_dim == 2 # ??

assert physical_dim == 2, "physical_dim must be 2 to ensure compatibility with the RBM implementation."

不用管他

Copilot · 2025-06-16T11:29:09Z

qmb/rbm.py

+        super().__init__()
+        self.sites: int = sites
+        assert physical_dim == 2  # ??
+        assert is_complex == True  # pylint: disable=singleton-comparison


Prefer assert is_complex over comparing to True to simplify the expression and satisfy linters without disabling them.

Suggested change

assert is_complex == True # pylint: disable=singleton-comparison

assert is_complex

不用管他

Copilot · 2025-06-16T11:29:10Z

qmb/rbm.py

+        self.dummy_param = torch.nn.Parameter(torch.empty(0))
+
+    @property
+    # A grammar sugar to determine device


[nitpick] Minor typo: change "grammar sugar" to the more standard term "syntactic sugar" for clarity.

Suggested change

# A grammar sugar to determine device

# A syntactic sugar to determine device

hzhangxyz · 2025-06-16T11:36:34Z

qmb/rbm.py

+        self.visible_dim = visible_dim
+        self.hidden_dim = hidden_dim
+        self.weights = torch.nn.Parameter(torch.Tensor(self.visible_dim, self.hidden_dim))
+        init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim).cuda())


你不应该在这里.cuda()，但这个copilot说的也不对，你应该直接

init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim)）

因为模型移动到不同的device（如gpu），是在model创建之后进行的。

hzhangxyz · 2025-06-16T11:37:45Z

qmb/rbm.py

+        super().__init__()
+        self.visible_dim = visible_dim
+        self.hidden_dim = hidden_dim
+        self.weights = torch.nn.Parameter(torch.Tensor(self.visible_dim, self.hidden_dim))


不要用torch.Tensor 这个东西， see: https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor

hzhangxyz · 2025-06-16T11:38:28Z

qmb/rbm.py

+            Probabilities.
+        """
+        e1 = (v @ self.visible_bias).view(v.size()[:-1])
+        mid2 = torch.nn.functional.linear(v, self.weights.T, self.hidden_bias)  # pylint: disable=E1102


给disable规则的具体名称，不要代号。

hzhangxyz · 2025-06-16T11:39:10Z

qmb/rbm.py

+    ) -> None:
+        super().__init__()
+        self.sites: int = sites
+        assert physical_dim == 2  # ??


不用管他

hzhangxyz · 2025-06-16T11:39:27Z

qmb/rbm.py

+        super().__init__()
+        self.sites: int = sites
+        assert physical_dim == 2  # ??
+        assert is_complex == True  # pylint: disable=singleton-comparison


不用管他

hzhangxyz · 2025-06-16T11:40:23Z

qmb/rbm.py

+        self.mlp_hidden_size: tuple[int, ...] = mpl_hidden_size
+
+        # Build Networks
+        self.probability: RBM = RBM(self.sites, self.rbm_hidden_dim)


你在这里并不需要给typing，这个他可以推导出来，如果你真的想给typing，请颗粒度一致，不要一个RBM一个torch.nn.Module。

hzhangxyz · 2025-06-16T11:41:04Z

qmb/rbm.py

+
+        Returns
+        -------
+        samples, amplitude


你这段注释并不符合numpy style。

修复了大部分小问题。

windy-pig force-pushed the dev/add-rbm-network branch from 5f08017 to fbd4a73 Compare June 12, 2025 15:26

windy-pig self-assigned this Jun 13, 2025

hzhangxyz requested a review from Copilot June 16, 2025 11:26

Copilot AI reviewed Jun 16, 2025

View reviewed changes

hzhangxyz reviewed Jun 16, 2025

View reviewed changes

windy-pig force-pushed the dev/add-rbm-network branch from 27651fc to 95ab358 Compare June 16, 2025 14:38

一个RBM网络实现，有较完整的文档和类型提示，

b52bc91

修复了大部分小问题。

windy-pig force-pushed the dev/add-rbm-network branch from 95ab358 to b52bc91 Compare June 16, 2025 15:09

hzhangxyz force-pushed the main branch from 3c7d134 to fd78860 Compare December 18, 2025 19:28

	init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim).cuda())
	init_range = 1.0 / torch.sqrt(torch.tensor(self.visible_dim, device=self.weights.device))

	assert physical_dim == 2 # ??
	assert physical_dim == 2, "physical_dim must be 2 to ensure compatibility with the RBM implementation."

	assert is_complex == True # pylint: disable=singleton-comparison
	assert is_complex

	# A grammar sugar to determine device
	# A syntactic sugar to determine device

Dev/add rbm network #35

Are you sure you want to change the base?

Dev/add rbm network #35

Uh oh!

Conversation

windy-pig commented Jun 12, 2025

Description

Checklist:

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Uh oh!

Copilot AI Jun 16, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jun 16, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jun 16, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jun 16, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants