File size: 7,366 Bytes
e6f08d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
import os 
import sys 
import numpy as np 
import torch 
import matplotlib 


matplotlib .use ("Agg")
import matplotlib .pyplot as plt 
import seaborn as sns 

from torch .utils .data import DataLoader 
from transformers import (
AutoTokenizer ,AutoModelForSequenceClassification ,DataCollatorWithPadding ,
)
from sklearn .metrics import (
classification_report ,confusion_matrix ,accuracy_score ,f1_score ,
)


BASE_DIR =os .path .dirname (os .path .dirname (os .path .abspath (__file__ )))
sys .path .append (os .path .join (BASE_DIR ,'training'))
from dataset import UrduTextDataset 

MAX_LENGTH =128 
BATCH_SIZE =16 














ROMAN_SENTIMENT_LABELS =["Positive","Negative","Neutral"]
URDU_SENTIMENT_LABELS =["Negative","Neutral","Positive"]
EMOTION_LABELS =["Joy","Anger","Fear","Sadness"]


def run_predictions (model ,dataset ,tokenizer ):
    """Run the model over the whole test set and return (predictions, true labels)."""


    collator =DataCollatorWithPadding (tokenizer =tokenizer )
    loader =DataLoader (dataset ,batch_size =BATCH_SIZE ,shuffle =False ,collate_fn =collator )

    model .eval ()
    all_preds =[]
    all_true =[]

    total_batches =len (loader )
    with torch .no_grad ():
        for i ,batch in enumerate (loader ,start =1 ):

            labels =batch .pop ("labels")
            logits =model (**batch ).logits 
            all_preds .append (logits .argmax (dim =-1 ).numpy ())
            all_true .append (labels .numpy ())

            if i %10 ==0 or i ==total_batches :
                print (f"      batch {i }/{total_batches }")

    return np .concatenate (all_preds ),np .concatenate (all_true )


def save_confusion_matrix (y_true ,y_pred ,label_names ,title ,out_path ):
    """Draw a confusion matrix heatmap (raw counts) and save it as a PNG."""
    cm =confusion_matrix (y_true ,y_pred ,labels =list (range (len (label_names ))))

    plt .figure (figsize =(7 ,6 ))
    sns .heatmap (
    cm ,annot =True ,fmt ='d',cmap ='Blues',
    xticklabels =label_names ,yticklabels =label_names ,cbar =False ,
    )
    plt .title (title )
    plt .xlabel ("Predicted label")
    plt .ylabel ("True label")
    plt .tight_layout ()
    plt .savefig (out_path ,dpi =150 )
    plt .close ()

    return cm 


def evaluate_task (task_name ,model_dir ,test_files ,task ,label_names ,results_dir ,plot_slug ):
    """Load one trained model, score it on its test set, save the plot, return a report string."""
    print (f"\n  Loading model + tokenizer from {os .path .relpath (model_dir ,BASE_DIR )}")
    tokenizer =AutoTokenizer .from_pretrained (model_dir )
    model =AutoModelForSequenceClassification .from_pretrained (model_dir )

    print ("  Loading test set...")
    test_dataset =UrduTextDataset (test_files ,tokenizer ,max_length =MAX_LENGTH ,task =task )
    print (f"  Test samples: {len (test_dataset )}")

    print ("  Running predictions...")
    y_pred ,y_true =run_predictions (model ,test_dataset ,tokenizer )


    accuracy =accuracy_score (y_true ,y_pred )
    macro_f1 =f1_score (y_true ,y_pred ,average ='macro',zero_division =0 )
    weighted_f1 =f1_score (y_true ,y_pred ,average ='weighted',zero_division =0 )


    report =classification_report (
    y_true ,y_pred ,
    labels =list (range (len (label_names ))),
    target_names =label_names ,
    digits =4 ,
    zero_division =0 ,
    )

    plot_path =os .path .join (results_dir ,f"{plot_slug }_confusion_matrix.png")
    cm =save_confusion_matrix (
    y_true ,y_pred ,label_names ,
    f"{task_name } - Confusion Matrix",plot_path ,
    )
    print (f"  Confusion matrix saved to {os .path .relpath (plot_path ,BASE_DIR )}")

    print (f"  Accuracy: {accuracy :.4f} | Macro F1: {macro_f1 :.4f} | Weighted F1: {weighted_f1 :.4f}")


    lines =[]
    lines .append ("="*70 )
    lines .append (f"  {task_name .upper ()}")
    lines .append ("="*70 )
    lines .append (f"Model directory : {os .path .relpath (model_dir ,BASE_DIR )}")
    lines .append ("Test files      : "+", ".join (os .path .basename (f )for f in test_files ))
    lines .append (f"Test samples    : {len (test_dataset )}")
    lines .append (f"Label order     : {label_names }")
    lines .append ("")
    lines .append (f"Accuracy        : {accuracy :.4f}")
    lines .append (f"Macro F1        : {macro_f1 :.4f}")
    lines .append (f"Weighted F1     : {weighted_f1 :.4f}")
    lines .append ("")
    lines .append ("Classification report:")
    lines .append (report )
    lines .append ("Confusion matrix (rows = true, columns = predicted):")
    header =" "*12 +"".join (f"{name :>12}"for name in label_names )
    lines .append (header )
    for name ,row in zip (label_names ,cm ):
        lines .append (f"{name :>12}"+"".join (f"{int (v ):>12}"for v in row ))
    lines .append ("")

    return "\n".join (lines )


def main ():
    print ("="*60 )
    print ("  Phase 5: Evaluating Sentiment & Emotion Models")
    print ("="*60 )

    torch .set_num_threads (os .cpu_count ()or 1 )

    data_dir =os .path .join (BASE_DIR ,'data')
    results_dir =os .path .join (BASE_DIR ,'results')
    os .makedirs (results_dir ,exist_ok =True )



    tasks =[
    {
    'task_name':"Sentiment Model - Roman Urdu test set",
    'model_dir':os .path .join (BASE_DIR ,'models','sentiment_model'),
    'test_files':[os .path .join (data_dir ,'roman_urdu_sentiment_test.csv')],
    'task':"sentiment",
    'label_names':ROMAN_SENTIMENT_LABELS ,
    'plot_slug':"sentiment_roman",
    },
    {
    'task_name':"Sentiment Model - Urdu script test set",
    'model_dir':os .path .join (BASE_DIR ,'models','sentiment_model'),
    'test_files':[os .path .join (data_dir ,'urdu_sentiment_corpus_test.csv')],
    'task':"sentiment",
    'label_names':URDU_SENTIMENT_LABELS ,
    'plot_slug':"sentiment_urdu_script",
    },
    {
    'task_name':"Emotion Model",
    'model_dir':os .path .join (BASE_DIR ,'models','emotion_model'),
    'test_files':[os .path .join (data_dir ,'semeval_emotion_test.csv')],
    'task':"emotion",
    'label_names':EMOTION_LABELS ,
    'plot_slug':"emotion",
    },
    ]

    all_reports =[]
    total_steps =len (tasks )+2 

    for step ,cfg in enumerate (tasks ,start =1 ):
        print (f"\n[{step }/{total_steps }] Evaluating: {cfg ['task_name']}")
        all_reports .append (evaluate_task (results_dir =results_dir ,**cfg ))

    print (f"\n[{total_steps -1 }/{total_steps }] Writing results.txt...")
    results_path =os .path .join (results_dir ,'results.txt')
    with open (results_path ,'w',encoding ='utf-8')as f :
        f .write ("Phase 5 Evaluation Results\n")
        f .write ("Urdu Sentiment & Emotion Analysis Engine (XLM-RoBERTa)\n\n")
        f .write ("NOTE ON SENTIMENT LABELS\n")
        f .write ("The two sentiment sources encode 0/1/2 differently (Positive and Negative\n")
        f .write ("are swapped between them), so they are scored separately, each under its own\n")
        f .write ("label names. See the comment block in evaluation/evaluate_models.py.\n\n")
        f .write ("\n".join (all_reports ))
    print (f"  Saved to {os .path .relpath (results_path ,BASE_DIR )}")

    print (f"\n[{total_steps }/{total_steps }] Done. Everything is in the results/ folder:")
    for name in sorted (os .listdir (results_dir )):
        print (f"  - results/{name }")


if __name__ =="__main__":
    main ()