nkululeko 0.95.2__py3-none-any.whl → 0.95.4__py3-none-any.whl

nkululeko/constants.py

@@ -1,2 +1,2 @@
-VERSION="0.95.2"
+VERSION="0.95.4"
 SAMPLING_RATE = 16000

nkululeko/data/dataset.py

@@ -6,7 +6,6 @@ from random import sample
 
 import numpy as np
 import pandas as pd
-
 import audformat
 
 from nkululeko.filter_data import DataFilter
@@ -905,6 +904,20 @@ class Dataset:
         mappings = self.util.config_val_data(self.name, "mapping", False)
         if mappings:
             mapping = ast.literal_eval(mappings)
+            # mapping should be a dictionary, the keys might encode lists.
+            keys = list(mapping.keys())
+            for key in keys:
+                # a comma in the key means that the key is a list of labels
+                if "," in key:
+                    # split the key and create a list
+                    key_list = [k.strip() for k in key.split(",")]
+                    # create a new mapping for each key
+                    for k in key_list:
+                        mapping[k] = mapping[key]
+                    # remove the old key
+                    del mapping[key]
+            if pd.api.types.is_numeric_dtype(df[target]):
+                df[target] = df[target].astype("string")
             df[target] = df[target].map(mapping)
             # remove any exiting nan values after mapping
             df = df.dropna()

nkululeko/experiment.py

@@ -23,7 +23,7 @@ from nkululeko.plots import Plots
 from nkululeko.reporting.report import Report
 from nkululeko.runmanager import Runmanager
 from nkululeko.scaler import Scaler
-from nkululeko.test_predictor import TestPredictor
+from nkululeko.testing_predictor import TestPredictor
 from nkululeko.utils.util import Util
 
 

nkululeko/modelrunner.py

@@ -250,14 +250,19 @@ class Modelrunner:
         balancing = self.util.config_val("FEATS", "balancing", False)
         if balancing:
             self.util.debug("Applying feature balancing using DataBalancer")
-            
-            # Initialize the data balancer
-            balancer = DataBalancer(random_state=42)
-            
+
+            # Get random state from config, fallback to 42 for backward compatibility
+            random_state = int(
+                self.util.config_val("FEATS", "balancing_random_state", 42)
+            )
+
+            # Initialize the data balancer with configurable random state
+            balancer = DataBalancer(random_state=random_state)
+
             # Apply balancing
             self.df_train, self.feats_train = balancer.balance_features(
                 df_train=self.df_train,
                 feats_train=self.feats_train,
                 target_column=self.target,
-                method=balancing
+                method=balancing,
             )

nkululeko/models/model_mlp_regression.py

@@ -174,6 +174,9 @@ class MLP_Reg_model(Model):
             logits = model(features.to(device)).reshape(-1)
             loss = self.criterion(logits, labels.to(device))
             # print(f'loss: {loss.item()}')
+            if torch.isnan(loss):
+                # possible that ccc returns NaN if batch contains only one value
+                continue
             losses.append(loss.item())
             optimizer.zero_grad()
             loss.backward()

nkululeko/nkuluflag.py

@@ -5,7 +5,7 @@ import os.path
 import sys
 
 from nkululeko.nkululeko import doit as nkulu
-from nkululeko.test import do_it as test_mod
+from nkululeko.testing import do_it as test_mod
 
 
 def doit(cla):