tgraphx 1.3.2__tar.gz → 1.3.3__tar.gz

{tgraphx-1.3.2/tgraphx.egg-info → tgraphx-1.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tgraphx
-Version: 1.3.2
+Version: 1.3.3
 Summary: Tensor-aware graph neural networks preserving spatial node feature layouts
 Author-email: Arash Sajjadi <arash.sajjadi@usask.ca>
 Maintainer-email: Arash Sajjadi <arash.sajjadi@usask.ca>

{tgraphx-1.3.2 → tgraphx-1.3.3}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "setuptools.build_meta"
 [project]
 name = "tgraphx"
 # Keep this in sync with tgraphx/__init__.py::__version__
-version = "1.3.2"
+version = "1.3.3"
 description = "Tensor-aware graph neural networks preserving spatial node feature layouts"
 readme = "README.md"
 requires-python = ">=3.10"

tgraphx-1.3.3/tests/test_reproducibility_easy_v133.py

@@ -0,0 +1,137 @@
+"""Reproducibility tests for Easy Mode (v1.3.3).
+
+Tests that:
+1. Synthetic data creation is seeded deterministically.
+2. NeighborLoader with same seed produces identical batches.
+3. CPU + deterministic=True gives exact same loss across two runs.
+4. Reproducibility state is recorded in result.config.
+5. CUDA (if available) at least produces finite results in deterministic mode.
+"""
+from __future__ import annotations
+
+import pytest
+import torch
+import tgraphx as tgx
+from tgraphx.reproducibility import set_seed
+
+
+# ── 1. Synthetic data reproducibility ────────────────────────────────────────
+
+
+class TestSyntheticDataReproducibility:
+    def test_same_seed_same_features(self):
+        d1 = tgx.easy.synthetic_tensor_node_classification(
+            num_nodes=32, node_shape=(4, 4, 4), num_classes=3, num_edges=100, seed=42,
+        )
+        d2 = tgx.easy.synthetic_tensor_node_classification(
+            num_nodes=32, node_shape=(4, 4, 4), num_classes=3, num_edges=100, seed=42,
+        )
+        assert torch.equal(d1.node_features, d2.node_features)
+        assert torch.equal(d1.edge_index, d2.edge_index)
+        assert torch.equal(d1.node_labels, d2.node_labels)
+
+    def test_different_seed_different_output(self):
+        d1 = tgx.easy.synthetic_tensor_node_classification(num_nodes=50, seed=0)
+        d2 = tgx.easy.synthetic_tensor_node_classification(num_nodes=50, seed=99)
+        # With different seeds the features should differ (with overwhelmingly high probability).
+        assert not torch.equal(d1.node_features, d2.node_features)
+
+
+# ── 2. NeighborLoader batch order reproducibility ────────────────────────────
+
+
+class TestNeighborLoaderReproducibility:
+    def test_same_seed_same_first_batch_seed_nodes(self):
+        from tgraphx import Graph, NeighborLoader
+        x = torch.randn(100, 8)
+        ei = torch.randint(0, 100, (2, 400))
+        y = torch.randint(0, 3, (100,))
+        g = Graph(node_features=x, edge_index=ei, y=y)
+
+        loader_a = NeighborLoader(g, fanouts=[5, 3], batch_size=8, shuffle=True, seed=7)
+        loader_b = NeighborLoader(g, fanouts=[5, 3], batch_size=8, shuffle=True, seed=7)
+
+        seeds_a = [b.seed_node_ids.clone() for b in loader_a]
+        seeds_b = [b.seed_node_ids.clone() for b in loader_b]
+        assert len(seeds_a) == len(seeds_b)
+        for a, b in zip(seeds_a, seeds_b):
+            assert torch.equal(a, b)
+
+
+# ── 3. Easy Mode CPU deterministic reproducibility ───────────────────────────
+
+
+class TestEasyModeDeterministicCPU:
+    def _run(self):
+        set_seed(42, deterministic=True)
+        data = tgx.easy.synthetic_tensor_node_classification(
+            num_nodes=32, node_shape=(4, 4, 4), num_classes=3, num_edges=100, seed=42,
+        )
+        return tgx.easy.train_node_classifier(
+            data, epochs=2, batch_size=8, fanouts=[4, 2],
+            verbose=False, seed=42, deterministic=True, device="cpu",
+        )
+
+    def test_deterministic_cpu_exact_match(self):
+        r1 = self._run()
+        r2 = self._run()
+        diff = abs(r1.metrics["loss"] - r2.metrics["loss"])
+        assert diff < 1e-7, f"CPU deterministic diff too large: {diff:.2e}"
+
+    def test_deterministic_flag_in_config(self):
+        r = self._run()
+        assert r.config.get("deterministic") is True
+
+    def test_reproducibility_state_recorded(self):
+        r = self._run()
+        state = r.config.get("reproducibility_state", {})
+        assert state.get("seed") == 42
+        assert state.get("deterministic") is True
+        assert "torch_version" in state
+        assert "cuda_available" in state
+
+    def test_default_non_deterministic_still_runs(self):
+        """Default deterministic=False must not crash."""
+        data = tgx.easy.synthetic_tensor_node_classification(
+            num_nodes=32, node_shape=(4, 4, 4), num_classes=2, num_edges=100, seed=0,
+        )
+        r = tgx.easy.train_node_classifier(
+            data, epochs=1, batch_size=8, fanouts=[3, 2],
+            verbose=False, seed=0, device="cpu",
+        )
+        assert "loss" in r.metrics
+        assert r.config.get("deterministic") is False
+
+
+# ── 4. set_seed return value ──────────────────────────────────────────────────
+
+
+class TestSetSeedReturnValue:
+    def test_returns_dict_with_required_keys(self):
+        state = set_seed(42, deterministic=True)
+        assert isinstance(state, dict)
+        assert state["seed"] == 42
+        assert state["deterministic"] is True
+        assert "torch_version" in state
+        assert "cuda_available" in state
+
+    def test_non_deterministic_mode(self):
+        state = set_seed(7, deterministic=False)
+        assert state["deterministic"] is False
+
+
+# ── 5. CUDA smoke (if available) ──────────────────────────────────────────────
+
+
+class TestEasyModeDeviceCUDA:
+    @pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA unavailable")
+    def test_deterministic_cuda_finite_loss(self):
+        set_seed(42, deterministic=True)
+        data = tgx.easy.synthetic_tensor_node_classification(
+            num_nodes=32, node_shape=(4, 4, 4), num_classes=3, num_edges=100, seed=42,
+        )
+        r = tgx.easy.train_node_classifier(
+            data, epochs=1, batch_size=8, fanouts=[4, 2],
+            verbose=False, seed=42, deterministic=True, device="cuda",
+        )
+        assert torch.isfinite(torch.tensor(r.metrics["loss"]))

{tgraphx-1.3.2 → tgraphx-1.3.3}/tgraphx/__init__.py

@@ -10,7 +10,7 @@ Common one-liner imports::
 """
 
 # Keep this in sync with [project].version in pyproject.toml.
-__version__ = "1.3.2"
+__version__ = "1.3.3"
 
 # ── Core data structures ──────────────────────────────────────────────────────
 from .core.graph import Graph, GraphBatch

{tgraphx-1.3.2 → tgraphx-1.3.3}/tgraphx/easy/_workflows.py

@@ -31,6 +31,7 @@ def train_node_classifier(
     verbose: bool = True,
     config: Optional[EasyConfig] = None,
     dashboard_dir: Optional[str] = None,
+    deterministic: bool = False,
 ) -> EasyResult:
     """Train a node classifier on a graph.
 
@@ -97,9 +98,10 @@ def train_node_classifier(
         )
 
     if seed is not None:
-        torch.manual_seed(seed)
         from tgraphx.reproducibility import set_seed
-        set_seed(seed)
+        reproducibility_state = set_seed(seed, deterministic=deterministic)
+    else:
+        reproducibility_state = {"seed": None, "deterministic": deterministic}
 
     dev = _resolve_device(device)
     if fanouts is None:
@@ -133,10 +135,12 @@ def train_node_classifier(
         "batch_size": batch_size,
         "device": str(dev),
         "seed": seed,
+        "deterministic": deterministic,
         "fanouts": fanouts,
         "hidden_channels": hidden_channels,
         "num_classes": num_classes,
         "node_shape": list(node_shape),
+        "reproducibility_state": reproducibility_state,
     }
 
     history: List[Dict[str, float]] = []

{tgraphx-1.3.2 → tgraphx-1.3.3/tgraphx.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tgraphx
-Version: 1.3.2
+Version: 1.3.3
 Summary: Tensor-aware graph neural networks preserving spatial node feature layouts
 Author-email: Arash Sajjadi <arash.sajjadi@usask.ca>
 Maintainer-email: Arash Sajjadi <arash.sajjadi@usask.ca>

{tgraphx-1.3.2 → tgraphx-1.3.3}/tgraphx.egg-info/SOURCES.txt

@@ -102,6 +102,7 @@ tests/test_random_walk_sample.py
 tests/test_readme_links.py
 tests/test_release_validation_v030.py
 tests/test_reproducibility.py
+tests/test_reproducibility_easy_v133.py
 tests/test_rl_baselines.py
 tests/test_rl_callbacks_v120.py
 tests/test_rl_high_level_api.py