nkululeko 0.85.0__py3-none-any.whl → 0.85.1__py3-none-any.whl

nkululeko/constants.py

@@ -1,2 +1,2 @@
-VERSION="0.85.0"
+VERSION="0.85.1"
 SAMPLING_RATE = 16000

nkululeko/modelrunner.py

@@ -151,9 +151,9 @@ class Modelrunner:
                 self.df_train, self.df_test, self.feats_train, self.feats_test
             )
         elif model_type == "finetune":
-            from nkululeko.models.model_tuned import Pretrained_model
+            from nkululeko.models.model_tuned import TunedModel
 
-            self.model = Pretrained_model(
+            self.model = TunedModel(
                 self.df_train, self.df_test, self.feats_train, self.feats_test
             )
         elif model_type == "gmm":

nkululeko/models/model_tuned.py

@@ -1,48 +1,39 @@
 """
-Code based on @jwagner
+Code based on @jwagner.
 """
 
-import audiofile
-import audeer
-import audmetric
-import datasets
-import pandas as pd
-import transformers
-from nkululeko.utils.util import Util
-import nkululeko.glob_conf as glob_conf
-from nkululeko.models.model import Model as BaseModel
-
-# import nkululeko.models.finetune_model as fm
-from nkululeko.reporting.reporter import Reporter
-import torch
-import ast
-import numpy as np
-from sklearn.metrics import recall_score
-from collections import OrderedDict
-import os
+import dataclasses
 import json
+import os
 import pickle
-import dataclasses
 import typing
 
+import datasets
+import numpy as np
+import pandas as pd
 import torch
 import transformers
-from transformers.models.wav2vec2.modeling_wav2vec2 import (
-    Wav2Vec2PreTrainedModel,
-    Wav2Vec2Model,
-)
+from transformers.models.wav2vec2.modeling_wav2vec2 import Wav2Vec2Model
+from transformers.models.wav2vec2.modeling_wav2vec2 import Wav2Vec2PreTrainedModel
 
+import audeer
+import audiofile
+import audmetric
+
+import nkululeko.glob_conf as glob_conf
+from nkululeko.models.model import Model as BaseModel
+from nkululeko.reporting.reporter import Reporter
 
-class Pretrained_model(BaseModel):
+
+class TunedModel(BaseModel):
 
     is_classifier = True
 
     def __init__(self, df_train, df_test, feats_train, feats_test):
-        """Constructor taking the configuration and all dataframes"""
+        """Constructor taking the configuration and all dataframes."""
         super().__init__(df_train, df_test, feats_train, feats_test)
-        super().set_model_type("ann")
+        super().set_model_type("finetuned")
         self.name = "finetuned_wav2vec2"
-        self.model_type = "finetuned"
         self.target = glob_conf.config["DATA"]["target"]
         labels = glob_conf.labels
         self.class_num = len(labels)
@@ -74,22 +65,11 @@ class Pretrained_model(BaseModel):
 
         for split in ["train", "dev"]:
             df = data_sources[split]
-            df[target_name] = df[target_name].astype("float")
-
-            y = pd.Series(
-                data=df.itertuples(index=False, name=None),
-                index=df.index,
-                dtype=object,
-                name="labels",
-            )
-
+            y = df[target_name].astype("float")
             y.name = "targets"
             df = y.reset_index()
             df.start = df.start.dt.total_seconds()
             df.end = df.end.dt.total_seconds()
-
-            # print(f"{split}: {len(df)}")
-
             ds = datasets.Dataset.from_pandas(df)
             dataset[split] = ds
 
@@ -143,12 +123,6 @@ class Pretrained_model(BaseModel):
     def set_model_type(self, type):
         self.model_type = type
 
-    def is_ann(self):
-        if self.model_type == "ann":
-            return True
-        else:
-            return False
-
     def set_testdata(self, data_df, feats_df):
         self.df_test, self.feats_test = data_df, feats_df
 
@@ -207,7 +181,8 @@ class Pretrained_model(BaseModel):
             "ACC": audmetric.accuracy,
         }
 
-        truth = p.label_ids[:, 0].astype(int)
+        # truth = p.label_ids[:, 0].astype(int)
+        truth = p.label_ids
         preds = p.predictions
         preds = np.argmax(preds, axis=1)
         scores = {}
@@ -216,8 +191,7 @@ class Pretrained_model(BaseModel):
         return scores
 
     def train(self):
-        """Train the model"""
-
+        """Train the model."""
         model_root = self.util.get_path("model_dir")
         log_root = os.path.join(self.util.get_exp_dir(), "log")
         audeer.mkdir(log_root)
@@ -225,16 +199,17 @@ class Pretrained_model(BaseModel):
         conf_file = os.path.join(self.torch_root, "config.json")
         if os.path.isfile(conf_file):
             self.util.debug(f"reusing finetuned model: {conf_file}")
-            self.load(self.run, self.epoch)
+            self.load(self.run, self.epoch_num)
             return
         targets = pd.DataFrame(self.dataset["train"]["targets"])
         counts = targets[0].value_counts().sort_index()
         train_weights = 1 / counts
         train_weights /= train_weights.sum()
-        # print(train_weights)
-        criterion_gender = torch.nn.CrossEntropyLoss(
+        self.util.debug("train weights: {train_weights}")
+        criterion = torch.nn.CrossEntropyLoss(
             weight=torch.Tensor(train_weights).to("cuda"),
         )
+        # criterion = torch.nn.CrossEntropyLoss()
 
         class Trainer(transformers.Trainer):
 
@@ -246,14 +221,12 @@ class Pretrained_model(BaseModel):
             ):
 
                 targets = inputs.pop("labels").squeeze()
-                targets_gender = targets.type(torch.long)
+                targets = targets.type(torch.long)
 
                 outputs = model(**inputs)
-                logits_gender = outputs[0].squeeze()
+                logits = outputs[0].squeeze()
 
-                loss_gender = criterion_gender(logits_gender, targets_gender)
-
-                loss = loss_gender
+                loss = criterion(logits, targets)
 
                 return (loss, outputs) if return_outputs else loss
 
@@ -325,7 +298,7 @@ class Pretrained_model(BaseModel):
             self.df_test[self.target].to_numpy().astype(float),
             predictions,
             self.run,
-            self.epoch,
+            self.epoch_num,
         )
         return report
 
@@ -371,13 +344,13 @@ class ModelOutput(transformers.file_utils.ModelOutput):
 
 class ModelHead(torch.nn.Module):
 
-    def __init__(self, config, num_labels):
+    def __init__(self, config):
 
         super().__init__()
 
         self.dense = torch.nn.Linear(config.hidden_size, config.hidden_size)
         self.dropout = torch.nn.Dropout(config.final_dropout)
-        self.out_proj = torch.nn.Linear(config.hidden_size, num_labels)
+        self.out_proj = torch.nn.Linear(config.hidden_size, config.num_labels)
 
     def forward(self, features, **kwargs):
 
@@ -398,7 +371,7 @@ class Model(Wav2Vec2PreTrainedModel):
         super().__init__(config)
 
         self.wav2vec2 = Wav2Vec2Model(config)
-        self.cat = ModelHead(config, 2)
+        self.cat = ModelHead(config)
         self.init_weights()
 
     def freeze_feature_extractor(self):

{nkululeko-0.85.0.dist-info → nkululeko-0.85.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.85.0
+Version: 0.85.1
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -333,6 +333,10 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.85.1
+--------------
+* fixed bug in model_finetuned that label_num was constant 2
+
 Version 0.85.0
 --------------
 * first version with finetuning wav2vec2 layers

{nkululeko-0.85.0.dist-info → nkululeko-0.85.1.dist-info}/RECORD

@@ -2,7 +2,7 @@ nkululeko/__init__.py,sha256=62f8HiEzJ8rG2QlTFJXUCMpvuH3fKI33DoJSj33mscc,63
 nkululeko/aug_train.py,sha256=YhuZnS_WVWnun9G-M6g5n6rbRxoVREz6Zh7k6qprFNQ,3194
 nkululeko/augment.py,sha256=4MG0apTAG5RgkuJrYEjGgDdbodZWi_HweSPNI1JJ5QA,3051
 nkululeko/cacheddataset.py,sha256=lIJ6hUo5LoxSrzXtWV8mzwO7wRtUETWnOQ4ws2XfL1E,969
-nkululeko/constants.py,sha256=flWSUNQs4r0X0SgoR1I72Mk49cRUdpBN8Zng8sySFBE,39
+nkululeko/constants.py,sha256=WnTSXQjJmWE-IrXcNSEa5FFV_83-z0EOGXa9trq00uE,39
 nkululeko/demo.py,sha256=8bl15Kitoesnz8oa8yrs52T6YCSOhWbbq9PnZ8Hj6D0,3232
 nkululeko/demo_feats.py,sha256=sAeGFojhEj9WEDFtG3SzPBmyYJWLF2rkbpp65m8Ujo4,2025
 nkululeko/demo_predictor.py,sha256=es56xbT8ifkS_vnrlb5NTZT54gNmeUtNlA4zVA_gnN8,4757
@@ -13,7 +13,7 @@ nkululeko/feature_extractor.py,sha256=8mssYKmo4LclVI-hiLmJEDZ0ZPyDavFG2YwtXcrGzw
 nkululeko/file_checker.py,sha256=LoLnL8aHpW-axMQ46qbqrManTs5otG9ShpEZuz9iRSk,3474
 nkululeko/filter_data.py,sha256=w-X2mhKdYr5DxDIz50E5yzO6Jmzk4jjDBoXsgOOVtcA,7222
 nkululeko/glob_conf.py,sha256=KL9YJQTHvTztxo1vr25qRRgaPnx4NTg0XrdbovKGMmw,525
-nkululeko/modelrunner.py,sha256=pPhvTh1rIrFQg5Ox9T1KoFJ4wRcLCmJl7LFud2DA41w,10464
+nkululeko/modelrunner.py,sha256=iCmfJxsS2UafcikjRdUqPQuqQMOYA-Ctr3et3HeNR3c,10452
 nkululeko/multidb.py,sha256=fG3VukEWP1vreVN4gB1IRXxwwg4jLftsSEYtu0o1f78,5634
 nkululeko/nkuluflag.py,sha256=PGWSmZz-PiiHLgcZJAoGOI_Y-sZDVI1ksB8p5r7riWM,3725
 nkululeko/nkululeko.py,sha256=Kn3s2E3yyH8cJ7z6lkMxrnqtCxTu7-qfe9Zr_ONTD5g,1968
@@ -89,7 +89,7 @@ nkululeko/models/model_svm.py,sha256=rsME3KvKvNG7bdE5lbvYUu85WZhaASZxxmdNDIVJRZ4
 nkululeko/models/model_svr.py,sha256=_YZeksqB3eBENGlg3g9RwYFlk9rQQ-XCeNBKLlGGVoE,725
 nkululeko/models/model_tree.py,sha256=rf16faUm4o2LJgkoYpeY998b8DQIvXZ73_m1IS3TnnE,417
 nkululeko/models/model_tree_reg.py,sha256=IgQcPTE-304HQLYSKPF8Z4ot_Ur9dH01fZjS0nXke_M,428
-nkululeko/models/model_tuned.py,sha256=zmagIE3QHP67_XJCx5r7ZXBojsp6SC8IS-L3XRWmCEk,15650
+nkululeko/models/model_tuned.py,sha256=WJplfUK3CGLSd2mahUrPSjMvqjPfxLp99KFeZaz2AbU,15098
 nkululeko/models/model_xgb.py,sha256=Thgx5ESdIok4v72mKh4plxpo4smGcKALWNCJTDScY0M,447
 nkululeko/models/model_xgr.py,sha256=aGBtNGLWjOE_2rICGYGFxmT8DtnHYsIl1lIpMtghHsY,418
 nkululeko/reporting/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -106,8 +106,8 @@ nkululeko/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nkululeko/utils/files.py,sha256=UiGAtZRWYjHSvlmPaTMtzyNNGE6qaLaxQkybctS7iRM,4021
 nkululeko/utils/stats.py,sha256=1yUq0FTOyqkU8TwUocJRYdJaqMU5SlOBBRUun9STo2M,2829
 nkululeko/utils/util.py,sha256=b1IHFucRNuF9Iyv5IJeK4AEg0Rga0xKG80UM5GWWdHA,13816
-nkululeko-0.85.0.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
-nkululeko-0.85.0.dist-info/METADATA,sha256=Zt3H0FmIXOJvzyLOI0aC8VfvjrdIkd4uNvb937luo_k,36499
-nkululeko-0.85.0.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
-nkululeko-0.85.0.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
-nkululeko-0.85.0.dist-info/RECORD,,
+nkululeko-0.85.1.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
+nkululeko-0.85.1.dist-info/METADATA,sha256=RonY9PdKyHjwYsZ3T9TgEs1JNnY1qbMdDr-Sp6kcCW8,36591
+nkululeko-0.85.1.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
+nkululeko-0.85.1.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
+nkululeko-0.85.1.dist-info/RECORD,,