Ruff

experiments/arxiv/compare_trigger_policies/pipeline_config.py

@@ -18,19 +18,19 @@
 )
 from modyn.config.schema.pipeline.evaluation.metrics import AccuracyMetricConfig, F1ScoreMetricConfig
 
-
 arxiv_bytes_parser_function = (
     "import torch\n"
     "import numpy as np\n"
-        "def bytes_parser_function(data: bytes) -> str:\n"
-        "    return str(data, 'utf8')"
+    "def bytes_parser_function(data: bytes) -> str:\n"
+    "    return str(data, 'utf8')"
 )
 arxiv_evaluation_transformer_function = (
     "import torch\n"
     "def evaluation_transformer_function(model_output: torch.Tensor) -> torch.Tensor:\n"
     "    return torch.argmax(model_output, dim=-1)\n"
 )
 
+
 def gen_pipeline_config(
     config_ref: str,
     trigger_config: TriggerConfig,
@@ -40,7 +40,9 @@ def gen_pipeline_config(
 ) -> ModynPipelineConfig:
     num_classes = 172
     return ModynPipelineConfig(
-        pipeline=Pipeline(name=config_ref, description="Arxiv pipeline for comparing trigger policies", version="0.0.1"),
+        pipeline=Pipeline(
+            name=config_ref, description="Arxiv pipeline for comparing trigger policies", version="0.0.1"
+        ),
         model=ModelConfig(id="ArticleNet", config={"num_classes": num_classes}),
         model_storage=PipelineModelStorageConfig(full_model_strategy=FullModelStrategy(name="PyTorchFullModel")),
         training=TrainingConfig(
@@ -87,7 +89,9 @@ def gen_pipeline_config(
                     dataloader_workers=1,
                     tokenizer="DistilBertTokenizerTransform",
                     metrics=[
-                        AccuracyMetricConfig(evaluation_transformer_function=arxiv_evaluation_transformer_function, topn=1),
+                        AccuracyMetricConfig(
+                            evaluation_transformer_function=arxiv_evaluation_transformer_function, topn=1
+                        ),
                         AccuracyMetricConfig(evaluation_transformer_function="", topn=2),
                         AccuracyMetricConfig(evaluation_transformer_function="", topn=5),
                         AccuracyMetricConfig(evaluation_transformer_function="", topn=10),

experiments/arxiv/compare_trigger_policies/run.py

@@ -3,8 +3,8 @@
 import pandas as pd
 
 from experiments.arxiv.compare_trigger_policies.pipeline_config import gen_pipeline_config
-from experiments.utils.models import Experiment
 from experiments.utils.experiment_runner import run_multiple_pipelines
+from experiments.utils.models import Experiment
 from modyn.config.schema.pipeline import (
     EvalHandlerConfig,
     ModynPipelineConfig,
@@ -27,9 +27,8 @@
 _FIRST_TIMESTAMP = int(pd.to_datetime("1995-01-01").timestamp())
 _LAST_TIMESTAMP = int(pd.to_datetime("2024-07-01").timestamp())
 
-def construct_slicing_eval_handler(
-    execution_time: EvalHandlerExecutionTime = "manual"
-) -> EvalHandlerConfig:
+
+def construct_slicing_eval_handler(execution_time: EvalHandlerExecutionTime = "manual") -> EvalHandlerConfig:
     return EvalHandlerConfig(
         name="slidingmatrix",
         execution_time=execution_time,
@@ -95,9 +94,10 @@ def construct_pipelines(experiment: Experiment) -> list[ModynPipelineConfig]:
         )
     ]
 
+
 PERIODIC_EVAL_INTERVAL = [("current", "13w")]  # total: 1/2y
 
-# pretrain/cold start can be chosen post fuction by just dropping evaluation info before a certain date
+# pretrain/cold start can be chosen post function by just dropping evaluation info before a certain date
 _EXPERIMENT_REFS: dict[int, Experiment] = {
     # -------------------------------------------------------------------------------- #
     #         1X: Baselines with PERIODIC_EVAL_INTERVAL, executed with cautious        #
@@ -109,8 +109,8 @@ def construct_pipelines(experiment: Experiment) -> list[ModynPipelineConfig]:
     10: Experiment(
         name="arxiv-baseline-time",
         eval_handlers=(
-            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual") +
-            construct_between_trigger_eval_handler("manual")
+            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual")
+            + construct_between_trigger_eval_handler("manual")
         ),
         time_triggers={
             schedule: TimeTriggerConfig(every=schedule, start_timestamp=_FIRST_TIMESTAMP)
@@ -122,8 +122,8 @@ def construct_pipelines(experiment: Experiment) -> list[ModynPipelineConfig]:
     11: Experiment(
         name="arxiv-baseline-dataamount",
         eval_handlers=(
-            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual") +
-            construct_between_trigger_eval_handler("manual")
+            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual")
+            + construct_between_trigger_eval_handler("manual")
         ),
         data_amount_triggers={
             f"{num_samples}": DataAmountTriggerConfig(num_samples=num_samples)

experiments/huffpost/compare_trigger_policies/pipeline_config.py

@@ -21,8 +21,8 @@
 hp_bytes_parser_function = (
     "import torch\n"
     "import numpy as np\n"
-        "def bytes_parser_function(data: bytes) -> str:\n"
-        "    return str(data, 'utf8')"
+    "def bytes_parser_function(data: bytes) -> str:\n"
+    "    return str(data, 'utf8')"
 )
 hp_evaluation_transformer_function = (
     "import torch\n"
@@ -40,7 +40,9 @@ def gen_pipeline_config(
 ) -> ModynPipelineConfig:
     num_classes = 42
     return ModynPipelineConfig(
-        pipeline=Pipeline(name=config_ref, description="Huffpost pipeline for comparing trigger policies", version="0.0.1"),
+        pipeline=Pipeline(
+            name=config_ref, description="Huffpost pipeline for comparing trigger policies", version="0.0.1"
+        ),
         model=ModelConfig(id="ArticleNet", config={"num_classes": num_classes}),
         model_storage=PipelineModelStorageConfig(full_model_strategy=FullModelStrategy(name="PyTorchFullModel")),
         training=TrainingConfig(
@@ -61,7 +63,9 @@ def gen_pipeline_config(
                             name="default",
                             algorithm="AdamW",
                             source="PyTorch",
-                            param_groups=[OptimizerParamGroup(module="model", config={"lr": 0.00002, "weight_decay": 0.01})],
+                            param_groups=[
+                                OptimizerParamGroup(module="model", config={"lr": 0.00002, "weight_decay": 0.01})
+                            ],
                         )
                     ],
                 )
@@ -94,7 +98,9 @@ def gen_pipeline_config(
                     dataloader_workers=1,
                     tokenizer="DistilBertTokenizerTransform",
                     metrics=[
-                        AccuracyMetricConfig(evaluation_transformer_function=hp_evaluation_transformer_function, topn=1),
+                        AccuracyMetricConfig(
+                            evaluation_transformer_function=hp_evaluation_transformer_function, topn=1
+                        ),
                         AccuracyMetricConfig(evaluation_transformer_function="", topn=2),
                         AccuracyMetricConfig(evaluation_transformer_function="", topn=5),
                         AccuracyMetricConfig(evaluation_transformer_function="", topn=10),

experiments/huffpost/compare_trigger_policies/run.py

@@ -2,8 +2,8 @@
 
 import pandas as pd
 
-from experiments.utils.models import Experiment
 from experiments.utils.experiment_runner import run_multiple_pipelines
+from experiments.utils.models import Experiment
 from modyn.config.schema.pipeline import (
     EvalHandlerConfig,
     ModynPipelineConfig,
@@ -21,15 +21,14 @@
 from modyn.config.schema.pipeline.trigger.simple.data_amount import DataAmountTriggerConfig
 from modyn.config.schema.pipeline.trigger.simple.time import TimeTriggerConfig
 from modynclient.config.schema.client_config import ModynClientConfig, Supervisor
+
 from .pipeline_config import gen_pipeline_config
 
 _FIRST_TIMESTAMP = int(pd.to_datetime("2012-01-28").timestamp())
 _LAST_TIMESTAMP = int(pd.to_datetime("2022-09-24").timestamp())  # last: dummy
 
 
-def construct_slicing_eval_handler(
-    execution_time: EvalHandlerExecutionTime = "manual"
-) -> EvalHandlerConfig:
+def construct_slicing_eval_handler(execution_time: EvalHandlerExecutionTime = "manual") -> EvalHandlerConfig:
     return EvalHandlerConfig(
         name="slidingmatrix",
         execution_time=execution_time,
@@ -66,6 +65,7 @@ def construct_periodic_eval_handlers(
         for (interval, fake_interval) in intervals
     ]
 
+
 def construct_between_trigger_eval_handler(execution_time: EvalHandlerExecutionTime = "manual") -> EvalHandlerConfig:
     return EvalHandlerConfig(
         name="full",
@@ -75,6 +75,7 @@ def construct_between_trigger_eval_handler(execution_time: EvalHandlerExecutionT
         datasets=["huffpost_kaggle_all"],  # train and test
     )
 
+
 def construct_pipelines(experiment: Experiment) -> list[ModynPipelineConfig]:
     return [
         gen_pipeline_config(
@@ -93,6 +94,7 @@ def construct_pipelines(experiment: Experiment) -> list[ModynPipelineConfig]:
         )
     ]
 
+
 # total: 14weeks -> ~4mths (with quarterly evaluations the intervals slightly overlap by 1 month)
 PERIODIC_EVAL_INTERVAL = [("current", "7w")]
 
@@ -107,8 +109,8 @@ def construct_pipelines(experiment: Experiment) -> list[ModynPipelineConfig]:
     10: Experiment(
         name="hp-baseline-time",
         eval_handlers=(
-            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual") +
-            construct_between_trigger_eval_handler("manual")
+            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual")
+            + construct_between_trigger_eval_handler("manual")
         ),
         time_triggers={
             schedule: TimeTriggerConfig(every=schedule, start_timestamp=_FIRST_TIMESTAMP)
@@ -120,8 +122,8 @@ def construct_pipelines(experiment: Experiment) -> list[ModynPipelineConfig]:
     11: Experiment(
         name="hp-baseline-dataamount",
         eval_handlers=(
-            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual") +
-            construct_between_trigger_eval_handler("manual")
+            construct_periodic_eval_handlers(intervals=PERIODIC_EVAL_INTERVAL, execution_time="manual")
+            + construct_between_trigger_eval_handler("manual")
         ),
         data_amount_triggers={
             f"{num_samples}": DataAmountTriggerConfig(num_samples=num_samples)

experiments/yearbook/compare_trigger_policies/pipeline_config.py

@@ -16,7 +16,11 @@
 )
 from modyn.config.schema.pipeline.evaluation.config import EvalDataConfig
 from modyn.config.schema.pipeline.evaluation.handler import EvalHandlerConfig
-from modyn.config.schema.pipeline.evaluation.metrics import AccuracyMetricConfig, F1ScoreMetricConfig, RocAucMetricConfig
+from modyn.config.schema.pipeline.evaluation.metrics import (
+    AccuracyMetricConfig,
+    F1ScoreMetricConfig,
+    RocAucMetricConfig,
+)
 from modyn.config.schema.pipeline.model_storage import FullModelStrategy
 from modyn.config.schema.pipeline.sampling.config import NewDataStrategyConfig
 
@@ -113,9 +117,7 @@ def gen_pipeline_config(
                             num_classes=2,
                             average="micro",
                         ),
-                        RocAucMetricConfig(
-                            evaluation_transformer_function=yb_evaluation_transformer_function_rocauc
-                        )
+                        RocAucMetricConfig(evaluation_transformer_function=yb_evaluation_transformer_function_rocauc),
                     ],
                 )
                 for yb_dataset_name in ["yearbook_all", "yearbook_train", "yearbook_test"]