Merge pull request #9 from pytorch-lumo/dev1

Weakly update 2023.03.13

.docstr.yaml

@@ -0,0 +1,21 @@
+paths: ./src/lumo/
+badge: ./images # Path
+exclude:
+
+verbose: 3 # int (0-4)
+skip_magic: True # Boolean
+skip_file_doc: True # Boolean
+skip_property: True
+skip_init: True # Boolean
+#skip_class_def: True # Boolean
+skip_private: True # Boolean
+follow_links: True # Boolean
+accept_empty: True # Boolean
+ignore_names_file: .*/test # regex
+#fail_under: 90 # int
+#percentage_only: True # Boolean
+ignore_patterns:
+  callbacks:
+    - ".*"
+  exphook:
+    - ".*"

.gitignore

@@ -165,4 +165,5 @@ lumo_temp
 
 wandb
 .lumorc.json
-docs/
+.lumorc.public.json
+docs/

README.md

@@ -3,6 +3,7 @@
 [![PyPI version](https://badge.fury.io/py/lumo.svg)](https://badge.fury.io/py/lumo)
 ![Python-Test](https://github.com/pytorch-lumo/lumo/actions/workflows/python-test.yml/badge.svg)
 [![license](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://github.com/Lightning-AI/lightning/blob/master/LICENSE)
+![Python-doc](./images/docstr_coverage_badge.svg)
 
 `lumo` is a light-weight library to help construct your experiment code, record your experiment results, especially in the field of deep learning.
 

examples/quick_start.py

@@ -18,10 +18,10 @@ def add(x):
 
 db = (
     DatasetBuilder()
-    .add_input("xs", torch.arange(-2500, 2500, dtype=torch.float).unsqueeze(1))
-    .add_input("ys", torch.arange(-2500, 2500, dtype=torch.float), transform=add)
-    .add_output("xs", "xs")
-    .add_output("ys", "ys")
+        .add_input("xs", torch.arange(-2500, 2500, dtype=torch.float).unsqueeze(1))
+        .add_input("ys", torch.arange(-2500, 2500, dtype=torch.float), transform=add)
+        .add_output("xs", "xs")
+        .add_output("ys", "ys")
 )
 
 loader = db.DataLoader(batch_size=params.batch_size, shuffle=True)
@@ -50,12 +50,11 @@ def add(x):
         meter.mean.loss = loss
         meter.sum.c = 1
         record.record(meter)
-        logger.inline(record.avg())
+        logger.inline(record.agg())
     logger.newline()
 
 test_data = -torch.arange(1000, dtype=torch.float).unsqueeze(1)
 res = model(test_data)
 
-
 test_ys = test_data * 2
 print((res.long() - test_ys.long()).sum())

pyproject.toml

@@ -41,6 +41,7 @@ omit = [
     'src/lumo/cli/*',
     'src/lumo/vis/*',
     'src/lumo/decorators/*',
+    'src/lumo/exp/agent.py',
     'src/lumo/analyse/*',
     'src/lumo/sketch/*',
     'src/lumo/core/record_backend/*',
@@ -73,6 +74,24 @@ exclude_lines = [
     "if 0:",
     "if __name__ == .__main__.:",
     "if TYPE_CHECKING:",
+    # some tricky object that can not well tested.
+    "except ImportError",
+    "pass",
+    "return None",
+    "break", #
+
+    # Hard to test
+    "class RecordAbort",
+    "class GitCommit",
+    "def summary_experiment",
+    "def plot",
+    "if torch.cuda.is_available()", # ignore cuda
+    "if is_dist():", # ignore distribution
+
+    # Deprecated method:
+    "def add_input_transform",
+    "def add_output_transform",
+    "raise StopIteration"
 ]
 
 [tool.coverage.html]

requirements.txt

@@ -14,4 +14,5 @@ dbrecord
 packaging
 pandas
 hydra-core
-tensorboard
+tensorboard
+filelock

setup.py

@@ -30,5 +30,8 @@ def extract_version():
         keywords='lumo',
         packages=find_packages('src'),
         entry_points={
+            'console_scripts': [
+                'lumo = lumo.cli:main',
+            ]
         },
     )

src/lumo/__init__.py

@@ -1,7 +1,7 @@
 """
 
 """
-__version__ = "0.14.6"
+__version__ = "0.15.0"
 
 from .core import Params, ParamsType, MetricType, Meter, Record, TrainStage, BaseParams
 

src/lumo/analyse/collect.py

@@ -68,14 +68,14 @@ def flatten_metric(df, *keys: str):
 
 def collect_table_rows(metric_root=None) -> pd.DataFrame:
     """Collect all table_row into a pandas.DataFrame"""
-    res = []
+    res = {}
     logger = Logger()
     exp_map = list_all_metrics(metric_root)
     for k, rows in exp_map.items():
         # append existing row metrics
         global_dic = PDict(os.path.join(metricroot(), f'{k}.dict.sqlite'))
-        for row in global_dic.values():
-            res.append(row)
+        for test_name, row in global_dic.items():
+            res[test_name] = row
 
         if len(rows) == 0:
             continue
@@ -94,10 +94,10 @@ def collect_table_rows(metric_root=None) -> pd.DataFrame:
                 continue
             global_dic[test_name] = row
             shutil.move(row_fn, os.path.join(os.path.dirname(row_fn), f'.{test_name}.pkl'))
-            res.append(row)
+            res[test_name] = row
         global_dic.flush()
 
-    return pd.DataFrame(res)
+    return pd.DataFrame(list(res.values()))
 
 
 def replac(df: pd.DataFrame):

src/lumo/analyse/condition.py

@@ -7,14 +7,17 @@
 
 
 def in_(ser, value):
+    """pandas operation"""
     return ser.apply(lambda x: x in value)
 
 
 def not_in_(ser, value):
+    """pandas operation"""
     return ser.apply(lambda x: x not in value)
 
 
 def first(ser, value):
+    """pandas operation"""
     return ser.duplicated(value) == False
 
 
@@ -101,16 +104,19 @@ def __repr__(self):
         return f'{self.name} {self.op} {self.value}'
 
     def in_(self, lis):
+        """condition of `in` operation"""
         self.op = 'in'
         self.value = set(lis)
         return self
 
     def not_in_(self, lis):
+        """condition of `.duplicated(value) == False` operation"""
         self.op = 'notin'
         self.value = set(lis)
         return self
 
     def first(self, value):
+        """condition of `not in` operation"""
         self.op = 'first'
         self.value = value
         return self
@@ -121,7 +127,8 @@ def first(self, value):
 
 def filter_by_condition(df: DataFrame, *condition: Compare) -> DataFrame:
     """
-    简化版的 pipeline，仅用于等式和不等式筛选以及额外支持的 in/not_in 等功能
+    fdf = filter_by_condition(df,C['contition'] == True, C['c2'] > 1, ... )
+
     Args:
         df: padnas.DataFrame instance
         *condition: list of `~Compare` instance

src/lumo/cli/__init__.py

@@ -0,0 +1,56 @@
+import fire
+
+
+def rerun(test_name, **kwarg):
+    """
+    rerun a test
+        lumo rerun <test_name>
+        lumo rerun <test_name> --device=0
+
+    Args:
+        test_name:
+
+    Returns:
+
+    """
+    from lumo.exp.finder import retrieval_experiment
+    exp = retrieval_experiment(test_name)
+    if exp is not None:
+        exp.rerun([f'--{k}={v}' for k, v in kwarg.items()])
+    else:
+        exit(1)
+
+
+def note(test_name, description):
+    """
+    Add note to a test:
+        lumo note <test_name> description ;
+
+    Args:
+        test_name:
+        description:
+
+    Returns:
+
+    """
+    print(f"Adding note '{description}' to {test_name}")
+
+
+def server(port=8080):
+    """
+
+    Args:
+        port:
+
+    Returns:
+
+    """
+    print(f"Starting server on port {port}")
+
+
+def main():
+    fire.Fire({
+        'rerun': rerun,
+        'note': note,
+        'server': server,
+    })

src/lumo/cli/__main__.py

@@ -0,0 +1,5 @@
+import fire
+from lumo.cli import main
+
+if __name__ == '__main__':
+    main()