Merge pull request #41 from ispras/frontend

Frontend
ispras · Dec 6, 2024 · 33a185a · 33a185a
2 parents 1bf8166 + 380a752
commit 33a185a
Show file tree

Hide file tree

Showing 38 changed files with 1,853 additions and 985 deletions.
diff --git a/data/multiple-graphs/custom/small/raw/.info b/data/multiple-graphs/custom/small/raw/.info
@@ -1,7 +1,7 @@
 {
  "count": 8,
  "nodes": [5, 4, 4, 8, 6, 7, 7, 9],
- "directed": false,
+ "directed": true,
  "node_attributes": {
   "names": [
    "a", "b"

diff --git a/experiments/user_dataset.py b/experiments/user_dataset.py
@@ -122,9 +122,31 @@ def simgnn():
     print("len =", len(gen_dataset))
 
 
+def nx_to_ptg_converter():
+    from aux.utils import GRAPHS_DIR
+    from base.dataset_converter import networkx_to_ptg
+    from base.datasets_processing import DatasetManager
+    import networkx as nx
+
+    nx_path = GRAPHS_DIR / 'networkx-graphs' / 'input' / 'reply_graph.edgelist'
+    nx_graph = nx.read_edgelist(nx_path)
+    nx_graph = nx.to_undirected(nx_graph)
+    ptg_graph = networkx_to_ptg(nx_graph)
+    if ptg_graph.x is None:
+        ptg_graph.x = torch.zeros((ptg_graph.num_nodes, 1))
+    if ptg_graph.y is None:
+        ptg_graph.y = torch.zeros(ptg_graph.num_nodes)
+        ptg_graph.y[0] = 1
+    ptg_dataset = UserLocalDataset('test_dataset_single', [ptg_graph])
+    gen_dataset = DatasetManager.register_torch_geometric_local(ptg_dataset)
+    print(len(gen_dataset))
+
+
 if __name__ == '__main__':
 
     # local()
-    converted_local()
+    # converted_local()
     # api()
     # simgnn()
+
+    nx_to_ptg_converter()
diff --git a/metainfo/optimizers_parameters.json b/metainfo/optimizers_parameters.json
@@ -1,10 +1,10 @@
 {"Adam":
 	{
 		"lr": ["learn rate", "float", 0.001, {"min": 0.0001, "step": 0.001}, "learning rate"],
-		"beta1": ["beta1", "float", 0.9, {}, "coefficient used for computing running averages of gradient and its square"],
-		"beta2": ["beta2", "float", 0.999, {}, "coefficient used for computing running averages of gradient and its square"],
-		"eps": ["Epsilon", "float", 0.00000001, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 5e-4, {}, "weight decay (L2 penalty)"],
+		"beta1": ["beta1", "float", 0.9, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"beta2": ["beta2", "float", 0.999, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"eps": ["Epsilon", "float", 0.00000001, {"min": 0}, "term added to the denominator to improve numerical stability"],
+		"weight_decay":  ["Weight decay (L2)", "float", 5e-4, {"min": 0}, "weight decay (L2 penalty)"],
 		"amsgrad": ["AMSGrad", "bool", false, {}, "whether to use the AMSGrad"],
 		"_technical_parameter":
 		{
@@ -17,22 +17,22 @@
 		"lr": ["learn rate", "float", 1.0, {"min": 0.0001, "step": 1}, "coefficient that scale delta before it is applied to the parameters"],
 		"rho": ["rho", "float", 0.9, {}, "coefficient used for computing a running average of squared gradients"],
 		"eps": ["Epsilon", "float", 1e-6, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0, {}, "weight decay (L2 penalty)"]
+		"weight_decay":  ["Weight decay (L2)", "float", 0, {"min": 0}, "weight decay (L2 penalty)"]
 	},
 	"Adagrad":
 	{
 		"lr": ["learn rate", "float", 0.01, {"min": 0.0001, "step": 0.01}, "learning rate"],
-		"lr_decay": ["lr decay", "float", 0, {}, "learning rate decay"],
-		"eps": ["Epsilon", "float", 1e-10, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0, {}, "weight decay (L2 penalty)"]
+		"lr_decay": ["lr decay", "float", 0, {"min": 0}, "learning rate decay"],
+		"eps": ["Epsilon", "float", 1e-10, {"min": 0}, "term added to the denominator to improve numerical stability"],
+		"weight_decay":  ["Weight decay (L2)", "float", 0, {"min": 0}, "weight decay (L2 penalty)"]
 	},
 	"AdamW":
 	{
 		"lr": ["learn rate", "float", 0.001, {"min": 0.0001, "step": 0.001}, "learning rate"],
-		"beta1": ["beta1", "float", 0.9, {}, "coefficient used for computing running averages of gradient and its square"],
-		"beta2": ["beta2", "float", 0.999, {}, "coefficient used for computing running averages of gradient and its square"],
-		"eps": ["Epsilon", "float", 1e-8, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0.01, {}, "weight decay coefficient"],
+		"beta1": ["beta1", "float", 0.9, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"beta2": ["beta2", "float", 0.999, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"eps": ["Epsilon", "float", 1e-8, {"min": 0}, "term added to the denominator to improve numerical stability"],
+		"weight_decay":  ["Weight decay (L2)", "float", 0.01, {"min": 0}, "weight decay coefficient"],
 		"amsgrad": ["AMSGrad", "bool", false, {}, "whether to use the AMSGrad"],
 		"maximize": ["maximize", "bool", false, {}, "maximize the params based on the objective, instead of minimizing"],
 		"_technical_parameter":
@@ -44,9 +44,9 @@
 	"SparseAdam":
 	{
 		"lr": ["learn rate", "float", 0.001, {"min": 0.0001, "step": 0.001}, "learning rate "],
-		"beta1": ["beta1", "float", 0.9, {}, "coefficients used for computing running averages of gradient and its square"],
-		"beta2": ["beta2", "float", 0.999, {}, "coefficients used for computing running averages of gradient and its square"],
-		"eps": ["Epsilon", "float", 1e-8, {}, "term added to the denominator to improve numerical stability"],
+		"beta1": ["beta1", "float", 0.9, {"min": 0}, "coefficients used for computing running averages of gradient and its square"],
+		"beta2": ["beta2", "float", 0.999, {"min": 0}, "coefficients used for computing running averages of gradient and its square"],
+		"eps": ["Epsilon", "float", 1e-8, {"min": 0}, "term added to the denominator to improve numerical stability"],
 		"_technical_parameter":
 		{
 			"parameters_grouping":  [[
@@ -56,10 +56,10 @@
 	"Adamax":
 	{
 		"lr": ["learn rate", "float", 0.002, {"min": 0.0001, "step": 0.001}, "learning rate"],
-		"beta1": ["beta1", "float", 0.9, {}, "coefficient used for computing running averages of gradient and its square"],
-		"beta2": ["beta2", "float", 0.999, {}, "coefficient used for computing running averages of gradient and its square"],
-		"eps": ["Epsilon", "float", 1e-8, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0.01, {}, "weight decay (L2 penalty)"],
+		"beta1": ["beta1", "float", 0.9, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"beta2": ["beta2", "float", 0.999, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"eps": ["Epsilon", "float", 1e-8, {"min": 0}, "term added to the denominator to improve numerical stability"],
+		"weight_decay":  ["Weight decay (L2)", "float", 0.01, {"min": 0}, "weight decay (L2 penalty)"],
 		"maximize": ["maximize", "bool", false, {}, "maximize the params based on the objective, instead of minimizing"],
 		"_technical_parameter":
 		{
@@ -73,7 +73,7 @@
 		"lambd": ["lambd", "float", 0.0001, {}, "decay term"],
 		"alpha": ["alpha", "float", 0.75, {}, "power for eta update"],
 		"t0": ["t0", "float", 1000000.0, {}, "point at which to start averaging"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0, {}, "weight decay (L2 penalty)"]
+		"weight_decay":  ["Weight decay (L2)", "float", 0, {"min": 0}, "weight decay (L2 penalty)"]
 	},
 	"LBFGS":
 	{
@@ -85,11 +85,11 @@
 	"NAdam":
 	{
 		"lr": ["learn rate", "float", 0.002, {"min": 0.0001, "step": 0.001}, "learning rate"],
-		"beta1": ["beta1", "float", 0.9, {}, "coefficient used for computing running averages of gradient and its square"],
-		"beta2": ["beta2", "float", 0.999, {}, "coefficient used for computing running averages of gradient and its square"],
-		"eps": ["Epsilon", "float", 1e-8, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0, {}, "weight decay (L2 penalty)"],
-		"momentum_decay":  ["Momentum decay", "float", 0.004, {}, "momentum momentum_decay"],
+		"beta1": ["beta1", "float", 0.9, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"beta2": ["beta2", "float", 0.999, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"eps": ["Epsilon", "float", 1e-8, {"min": 0}, "term added to the denominator to improve numerical stability"],
+		"weight_decay":  ["Weight decay (L2)", "float", 0, {"min": 0}, "weight decay (L2 penalty)"],
+		"momentum_decay":  ["Momentum decay", "float", 0.004, {"min": 0}, "momentum momentum_decay"],
 		"_technical_parameter":
 		{
 			"parameters_grouping":  [[
@@ -99,10 +99,10 @@
 	"RAdam":
 	{
 		"lr": ["learn rate", "float", 0.001, {"min": 0.0001, "step": 0.001}, "learning rate"],
-		"beta1": ["beta1", "float", 0.9, {}, "coefficient used for computing running averages of gradient and its square"],
-		"beta2": ["beta2", "float", 0.999, {}, "coefficient used for computing running averages of gradient and its square"],
-		"eps": ["Epsilon", "float", 1e-8, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0, {}, "weight decay (L2 penalty)"],
+		"beta1": ["beta1", "float", 0.9, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"beta2": ["beta2", "float", 0.999, {"min": 0}, "coefficient used for computing running averages of gradient and its square"],
+		"eps": ["Epsilon", "float", 1e-8, {"min": 0}, "term added to the denominator to improve numerical stability"],
+		"weight_decay":  ["Weight decay (L2)", "float", 0, {"min": 0}, "weight decay (L2 penalty)"],
 		"_technical_parameter":
 		{
 			"parameters_grouping":  [[
@@ -112,9 +112,9 @@
 	"RMSprop":
 	{
 		"lr": ["learn rate", "float", 0.01, {"min": 0.0001, "step": 0.01}, "learning rate"],
-		"alpha": ["alpha", "float", 0.99, {}, "smoothing constant"],
-		"eps": ["Epsilon", "float", 1e-8, {}, "term added to the denominator to improve numerical stability"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0, {}, "weight decay (L2 penalty)"],
+		"alpha": ["alpha", "float", 0.99, {"min": 0}, "smoothing constant"],
+		"eps": ["Epsilon", "float", 1e-8, {"min": 0}, "term added to the denominator to improve numerical stability"],
+		"weight_decay":  ["Weight decay (L2)", "float", 0, {"min": 0}, "weight decay (L2 penalty)"],
 		"momentum":  ["momentum", "float", 0, {}, "momentum factor"],
 		"centered": ["centered", "bool", false, {}, "if True, compute the centered RMSProp, the gradient is normalized by an estimation of its variance"]
 	},
@@ -136,8 +136,8 @@
 	"SGD":
 	{
 		"lr": ["learn rate", "float", 0.001, {"min": 0.0001, "step": 0.001}, "learning rate"],
-		"weight_decay":  ["Weight decay (L2)", "float", 0, {}, "weight decay (L2 penalty)"],
-		"momentum":  ["momentum", "float", 0, {}, "momentum factor"],
+		"weight_decay":  ["Weight decay (L2)", "float", 0, {"min": 0}, "weight decay (L2 penalty)"],
+		"momentum":  ["momentum", "float", 0, {"min": 0}, "momentum factor"],
 		"dampening":  ["dampening", "float", 0, {}, "dampening for momentum"],
 		"nesterov": ["nesterov", "bool", false, {}, "enables Nesterov momentum"],
 		"maximize": ["maximize", "bool", false, {}, "maximize the params based on the objective, instead of minimizing"]

diff --git a/requirements2.txt b/requirements2.txt
@@ -1,7 +1,7 @@
 ## These are reqs needed for documentation
 
 numpy==1.26.3
-multiprocess
+multiprocess==0.70.16
 pydantic
 tqdm
 pyparsing
@@ -23,4 +23,4 @@ opt-einsum
 pandas==2.2.0
 pylint
 
-torch_geometric==2.3.1
+torch_geometric==2.3.1
diff --git a/src/aux/configs.py b/src/aux/configs.py
@@ -199,7 +199,7 @@ def set_defaults_config_pattern_info(
 
     def to_json(
             self
-    ):
+    ) -> dict:
         """ Special method which allows to use json.dumps() on Config object """
         return self.to_dict()
 

diff --git a/src/base/custom_datasets.py b/src/base/custom_datasets.py
@@ -36,41 +36,41 @@ def __init__(
     @property
     def node_attributes_dir(
             self
-    ):
+    ) -> Path:
         """ Path to dir with node attributes. """
         return self.root_dir / 'raw' / (self.name + '.node_attributes')
 
     @property
     def edge_attributes_dir(
             self
-    ):
+    ) -> Path:
         """ Path to dir with edge attributes. """
         return self.root_dir / 'raw' / (self.name + '.edge_attributes')
 
     @property
     def labels_dir(
             self
-    ):
+    ) -> Path:
         """ Path to dir with labels. """
         return self.root_dir / 'raw' / (self.name + '.labels')
 
     @property
     def edges_path(
             self
-    ):
+    ) -> Path:
         """ Path to file with edge list. """
         return self.root_dir / 'raw' / (self.name + '.ij')
 
     @property
     def edge_index_path(
             self
-    ):
+    ) -> Path:
         """ Path to dir with labels. """
         return self.root_dir / 'raw' / (self.name + '.edge_index')
 
     def check_validity(
             self
-    ):
+    ) -> None:
         """ Check that dataset files (graph and attributes) are valid and consistent with .info.
         """
         # Assuming info is OK
@@ -159,6 +159,7 @@ def build(
             return
 
         self.dataset_var_data = None
+        self.stats.update_var_config()
         self.dataset_var_config = dataset_var_config
         self.dataset = LocalDataset(self.results_dir, process_func=self._create_ptg)
 
@@ -178,6 +179,7 @@ def _compute_stat(
                     with open(self.node_attributes_dir / a, 'r') as f:
                         attr_node_attrs[a] = json.load(f)
 
+                # FIXME misha - for single graph [0]
                 edges = self.edge_index
                 node_map = (lambda i: str(self.node_map[i])) if self.node_map else lambda i: str(i)
 
@@ -223,8 +225,8 @@ def _compute_stat(
                         pearson_corr[i][j] = min(1, max(-1, pc))
 
                 return {'attributes': attrs, 'correlations': pearson_corr.tolist()}
-        else:
-            return super()._compute_stat(stat)
+
+        raise NotImplementedError()
 
     def _compute_dataset_data(
             self
@@ -420,7 +422,7 @@ def _iter_nodes(
 
     def _labeling_tensor(
             self,
-            g_ix=None
+            g_ix: int = None
     ) -> list:
         """ Returns list of labels (not tensors) """
         y = []
@@ -445,7 +447,7 @@ def _labeling_tensor(
 
     def _feature_tensor(
             self,
-            g_ix=None
+            g_ix: int = None
     ) -> list:
         """ Returns list of features (not tensors) for graph g_ix.
         """