Spaces:
Runtime error
Runtime error
Upload model_evaluation.py
Browse files- model_evaluation.py +173 -0
model_evaluation.py
ADDED
@@ -0,0 +1,173 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
from datasets import Dataset
|
3 |
+
from transformers import pipeline, GPT2Tokenizer
|
4 |
+
from sentence_transformers import SentenceTransformer, util
|
5 |
+
|
6 |
+
# Define paths and models
|
7 |
+
filename = "output_country_details.txt"
|
8 |
+
retrieval_model_name = 'output/sentence-transformer-finetuned/' #using a prefine-tuned model
|
9 |
+
gpt2_model_name = "gpt2"
|
10 |
+
csv_file_path = "train_dataset.csv"
|
11 |
+
output_csv_file_path = "updated_train_dataset.csv"
|
12 |
+
val_csv_file_path = "val_dataset.csv"
|
13 |
+
output_val_csv_file_path = "updated_val_csv.csv"
|
14 |
+
|
15 |
+
tokenizer = GPT2Tokenizer.from_pretrained(gpt2_model_name)
|
16 |
+
|
17 |
+
# Initialize models
|
18 |
+
try:
|
19 |
+
retrieval_model = SentenceTransformer(retrieval_model_name)
|
20 |
+
gpt_model = pipeline("text-generation", model=gpt2_model_name)
|
21 |
+
print("Models loaded successfully.")
|
22 |
+
except Exception as e:
|
23 |
+
print(f"Failed to load models: {e}")
|
24 |
+
|
25 |
+
def load_and_preprocess_text(filename):
|
26 |
+
"""
|
27 |
+
Load and preprocess text data from a file.
|
28 |
+
|
29 |
+
Parameters:
|
30 |
+
- filename (str): Path to the text file.
|
31 |
+
|
32 |
+
Returns:
|
33 |
+
- list[str]: A list of preprocessed text segments.
|
34 |
+
"""
|
35 |
+
try:
|
36 |
+
with open(filename, 'r', encoding='utf-8') as file:
|
37 |
+
segments = [line.strip() for line in file if line.strip()]
|
38 |
+
print("Text loaded and preprocessed successfully.")
|
39 |
+
return segments
|
40 |
+
except Exception as e:
|
41 |
+
print(f"Failed to load or preprocess text: {e}")
|
42 |
+
return []
|
43 |
+
|
44 |
+
segments = load_and_preprocess_text(filename)
|
45 |
+
|
46 |
+
def find_relevant_segment(user_query, segments):
|
47 |
+
"""
|
48 |
+
Find the most relevant text segment based on a user query.
|
49 |
+
|
50 |
+
Parameters:
|
51 |
+
- user_query (str): The user's query.
|
52 |
+
- segments (list[str]): List of text segments to search within.
|
53 |
+
|
54 |
+
Returns:
|
55 |
+
- str: The most relevant text segment.
|
56 |
+
"""
|
57 |
+
try:
|
58 |
+
query_embedding = retrieval_model.encode(user_query)
|
59 |
+
segment_embeddings = retrieval_model.encode(segments)
|
60 |
+
similarities = util.pytorch_cos_sim(query_embedding, segment_embeddings)[0]
|
61 |
+
best_idx = similarities.argmax()
|
62 |
+
return segments[best_idx]
|
63 |
+
except Exception as e:
|
64 |
+
print(f"Error finding relevant segment: {e}")
|
65 |
+
return ""
|
66 |
+
|
67 |
+
def generate_response(question):
|
68 |
+
"""
|
69 |
+
Generate a response to a given question by finding a relevant text segment and
|
70 |
+
using it to generate a more complete answer.
|
71 |
+
|
72 |
+
Parameters:
|
73 |
+
- question (str): The user's question.
|
74 |
+
|
75 |
+
Returns:
|
76 |
+
- str: Generated response.
|
77 |
+
"""
|
78 |
+
relevant_segment = find_relevant_segment(question, segments)
|
79 |
+
return generate_response_with_context(question, relevant_segment)
|
80 |
+
|
81 |
+
def generate_response_with_context(user_query, relevant_segment):
|
82 |
+
"""
|
83 |
+
Generate a response based on a user query and a relevant segment.
|
84 |
+
|
85 |
+
Parameters:
|
86 |
+
- user_query (str): The user's query.
|
87 |
+
- relevant_segment (str): A relevant fact or detail.
|
88 |
+
|
89 |
+
Returns:
|
90 |
+
- str: Formatted response incorporating the relevant segment.
|
91 |
+
"""
|
92 |
+
try:
|
93 |
+
prompt = f"Thank you for your question! Here is an additional fact about your topic: {relevant_segment}"
|
94 |
+
max_tokens = len(tokenizer(prompt)['input_ids']) + 50
|
95 |
+
response = gpt_model(prompt, max_length=max_tokens, temperature=0.25)[0]['generated_text']
|
96 |
+
return clean_up_response(response, relevant_segment)
|
97 |
+
except Exception as e:
|
98 |
+
print(f"Error generating response: {e}")
|
99 |
+
return ""
|
100 |
+
|
101 |
+
def clean_up_response(response, segment):
|
102 |
+
"""
|
103 |
+
Clean up the generated response to ensure it is tidy and presentable.
|
104 |
+
|
105 |
+
Parameters:
|
106 |
+
- response (str): The initial response generated by the model.
|
107 |
+
- segment (str): The segment used to generate the response.
|
108 |
+
|
109 |
+
Returns:
|
110 |
+
- str: A cleaned and formatted response.
|
111 |
+
"""
|
112 |
+
sentences = response.split('.')
|
113 |
+
cleaned_sentences = [sentence.strip() for sentence in sentences if sentence.strip() and sentence.strip() not in segment]
|
114 |
+
cleaned_response = '. '.join(cleaned_sentences).strip()
|
115 |
+
if cleaned_response and not cleaned_response.endswith((".", "!", "?")):
|
116 |
+
cleaned_response += "."
|
117 |
+
return cleaned_response
|
118 |
+
|
119 |
+
def process_dataset(csv_file_path, output_csv_file_path):
|
120 |
+
"""
|
121 |
+
Process the dataset by generating responses and evaluating their similarities.
|
122 |
+
|
123 |
+
Parameters:
|
124 |
+
- csv_file_path (str): Path to the CSV file containing the dataset.
|
125 |
+
- output_csv_file_path (str): Path where the updated dataset will be saved.
|
126 |
+
|
127 |
+
Prints:
|
128 |
+
- Path to the saved results and the average similarity score.
|
129 |
+
"""
|
130 |
+
df = pd.read_csv(csv_file_path)
|
131 |
+
dataset = Dataset.from_pandas(df)
|
132 |
+
updated_dataset = add_model_answers(dataset)
|
133 |
+
similarities = evaluate_similarity(updated_dataset)
|
134 |
+
updated_dataset = updated_dataset.add_column("similarity", similarities)
|
135 |
+
results_df = updated_dataset.to_pandas()
|
136 |
+
results_df.to_csv(output_csv_file_path, index=False)
|
137 |
+
average_similarity = sum(similarities) / len(similarities) if similarities else 0
|
138 |
+
print(f"Results saved to {output_csv_file_path}")
|
139 |
+
print(f"Average Similarity Score: {average_similarity:.3f}")
|
140 |
+
|
141 |
+
def add_model_answers(dataset):
|
142 |
+
"""
|
143 |
+
Add generated answers to the dataset.
|
144 |
+
|
145 |
+
Parameters:
|
146 |
+
- dataset (datasets.Dataset): The Hugging Face dataset object.
|
147 |
+
|
148 |
+
Returns:
|
149 |
+
- datasets.Dataset: Updated dataset with added answers.
|
150 |
+
"""
|
151 |
+
answers = [generate_response(q) for q in dataset['Question']]
|
152 |
+
dataset = dataset.add_column("Answer", answers)
|
153 |
+
return dataset
|
154 |
+
|
155 |
+
def evaluate_similarity(dataset):
|
156 |
+
"""
|
157 |
+
Evaluate the similarity of generated answers against ground truth answers.
|
158 |
+
|
159 |
+
Parameters:
|
160 |
+
- dataset (datasets.Dataset): The dataset containing both answers and ground truths.
|
161 |
+
|
162 |
+
Returns:
|
163 |
+
- list[float]: List of similarity scores.
|
164 |
+
"""
|
165 |
+
similarities = [util.pytorch_cos_sim(retrieval_model.encode(ans), retrieval_model.encode(gt))[0][0].item()
|
166 |
+
for ans, gt in zip(dataset['Answer'], dataset['GroundTruth'])]
|
167 |
+
return similarities
|
168 |
+
|
169 |
+
# Process datasets
|
170 |
+
process_dataset(csv_file_path, output_csv_file_path)
|
171 |
+
process_dataset(val_csv_file_path, output_val_csv_file_path)
|
172 |
+
|
173 |
+
|