feat: support nest log (#808)

* add flatten

* update test

swanlab/data/run/main.py

@@ -309,6 +309,17 @@ def config(self) -> SwanLabConfig:
         """
         return self.__config
 
+    def __flatten_dict(self, d: dict, parent_key='', sep='.') -> dict:
+        """Helper method to flatten nested dictionaries with dot notation"""
+        items = []
+        for k, v in d.items():
+            new_key = f"{parent_key}{sep}{k}" if parent_key else k
+            if isinstance(v, dict):
+                items.extend(self.__flatten_dict(v, new_key, sep=sep).items())
+            else:
+                items.append((new_key, v))
+        return dict(items)
+
     def log(self, data: dict, step: int = None):
         """
         Log a row of data to the current run. Unlike `swanlab.log`, this api will be called directly based on the
@@ -320,7 +331,8 @@ def log(self, data: dict, step: int = None):
         data : Dict[str, DataType]
             Data must be a dict.
             The key must be a string with 0-9, a-z, A-Z, " ", "_", "-", "/".
-            The value must be a `float`, `float convertible object`, `int` or `swanlab.data.BaseType`.
+            The value must be a `float`, `float convertible object`, `int`, `swanlab.data.BaseType`, or a nested dict.
+            For nested dicts, keys will be joined with dots (e.g., {'a': {'b': 1}} becomes {'a.b': 1}).
         step : int, optional
             The step number of the current data, if not provided, it will be automatically incremented.
             If step is duplicated, the data will be ignored.
@@ -347,15 +359,18 @@ def log(self, data: dict, step: int = None):
             )
             step = None
 
+        # 展平嵌套字典
+        flattened_data = self.__flatten_dict(data)
+
         log_return = {}
         # 遍历data，记录data
-        for k, v in data.items():
+        for k, v in flattened_data.items():
             _k = k
             k = check_key_format(k, auto_cut=True)
             if k != _k:
                 # 超过255字符，截断
                 swanlog.warning(f"Key {_k} is too long, cut to 255 characters.")
-                if k in data.keys():
+                if k in flattened_data.keys():
                     raise ValueError(f'tag: Not supported too long Key "{_k}" and auto cut failed')
             # ---------------------------------- 包装数据 ----------------------------------
             # 输入为可转换为float的数据类型

test/unit/data/run/test_main.py

@@ -113,13 +113,14 @@ def test_log_column_with_id(self):
     # ---------------------------------- 解析log数字/Line ----------------------------------
     def test_log_number_ok(self):
         run = SwanLabRun()
-        data = {"a": 1, "b": 0.1, "math.nan": math.nan, "math.inf": math.inf}
+        data = {"a": 1, "b": 0.1, "c": {"d": 2}, "math.nan": math.nan, "math.inf": math.inf}
         ll = run.log(data)
-        assert len(ll) == 4
+        assert len(ll) == 5
         # 都没有错误
         assert all([ll[k].is_error is False for k in ll])
         assert ll["a"].data == 1
         assert ll["b"].data == 0.1
+        assert ll["c.d"].data == 2
         assert ll["math.nan"].data == Line.nan
         assert ll["math.inf"].data == Line.inf
         assert all([ll[k].column_info.chart_type == ll[k].column_info.chart_type.LINE for k in ll])
@@ -146,9 +147,9 @@ def test_log_number_use_line(self):
         使用Line类型log，本质上应该与数字类型一样，数字类型是Line类型的语法糖
         """
         run = SwanLabRun()
-        data = {"a": Line(1), "b": Line(0.1), "math.nan": Line(math.nan), "math.inf": Line(math.inf)}
+        data = {"a": Line(1), "b": Line(0.1), "c": {"d": Line(2)}, "math.nan": Line(math.nan), "math.inf": Line(math.inf)}
         ll = run.log(data)
-        assert len(ll) == 4
+        assert len(ll) == 5
         # line(1)和[line(1)]是一样的
         ll2 = run.log({"a": [Line(1)]})
         assert ll2["a"].data == ll["a"].data