result table formatting and tabulate dependency

code/experiments/generate_tables.py

@@ -98,7 +98,7 @@ def process_betti(result_csv_prefix: str):
 
 
 def process():
-    results_csv_prefix = "ignore_temp_"
+    results_csv_prefix = "results_"
 
     res_betti = process_betti(results_csv_prefix)
     res_name_or = process_name_orientability(results_csv_prefix)

code/experiments/utils/results_processing/per_model.py

@@ -4,7 +4,7 @@
 from models.models import ModelType
 from datasets.transforms import TransformType
 import numpy as np
-from .utils import get_matching_indeces, get_metric_col_names, get_result_path
+from .utils import get_matching_indeces, get_metric_col_names, get_result_path, format_res_val
 
 
 def reduce(
@@ -40,10 +40,8 @@ def reduce(
                     get_matching_indeces(df, model_type, transform_type)
                 ]
                 metric_results.append(filtered_results[metric].max())
-            row[metric[5:]] = (
-                f"{np.max(metric_results):.4f} (std:{np.std(metric_results):.4f})"
-            )
-
+            row[metric[5:]] = format_res_val(np.max(metric_results), np.std(metric_results))
+
         new_row_df = pd.DataFrame([row])
         concat_df = (
             [df_results, new_row_df] if len(df_results) > 0 else [new_row_df]

code/experiments/utils/results_processing/per_task.py

@@ -4,7 +4,7 @@
 from models.models import ModelType
 from datasets.transforms import TransformType
 import numpy as np
-from .utils import get_matching_indeces, get_metric_col_names, get_result_path
+from .utils import get_matching_indeces, get_metric_col_names, get_result_path, format_res_val
 
 
 def reduce(
@@ -36,7 +36,6 @@ def reduce(
             ]
             max_ = filtered_results[metric].max()
             model_results.append(max_)
-
         metric_res_list.append(np.max(model_results))
     return metric_res_list
 
@@ -84,7 +83,7 @@ def per_task(
                 )
 
             if metric != "test_loss":
-                row = {"Metric": metric, "Mean": np.mean(res_for_metric)}
+                row = {"Metric": metric, "Mean": format_res_val(np.mean(res_for_metric))}
                 new_row_df = pd.DataFrame([row])
                 concat_df = (
                     [df_results, new_row_df]

code/experiments/utils/results_processing/per_transform.py

@@ -4,7 +4,7 @@
 from models.models import ModelType
 from datasets.transforms import TransformType
 import numpy as np
-from .utils import get_matching_indeces, get_metric_col_names, get_result_path
+from .utils import get_matching_indeces, get_metric_col_names, get_result_path, format_res_val
 
 
 def reduce(
@@ -37,9 +37,7 @@ def reduce(
             ]
             results.append(filtered_results[metric].max())
 
-        row[transform_type.name.lower()] = (
-            f"{np.max(results):.4f} (std:{np.std(results):.4f})"
-        )
+        row[transform_type.name.lower()] = format_res_val(np.max(results), np.std(results))
 
     return row
 

code/experiments/utils/results_processing/utils.py

@@ -1,9 +1,16 @@
-from typing import List, Tuple
+from typing import List, Tuple, Optional
 import pandas as pd
 from metrics.tasks import TaskType
 from models.models import ModelType
 from datasets.transforms import TransformType
 
+def format_res_val(
+    value: float, std: Optional[float] = None
+):
+    if std is None:
+        return f"{value:.2f}"
+    else: 
+        return f"{value:.2f} ({std:.2f} SD)"    
 
 def get_matching_indeces(
     df: pd.DataFrame, model_type: ModelType, transform_type: TransformType

pyproject.toml

@@ -35,6 +35,7 @@ torchvision = "0.18.0"
 gensim = "4.3.2"
 omegaconf = "2.3.0"
 pydantic-settings = "2.2.1"
+tabulate = "0.9.0"
 
 [tool.poetry.dev-dependencies]
 black = {version = "^21.12b0", allow-prereleases = true}