Datasets:
Tasks:
Text2Text Generation
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
< 1K
ArXiv:
Tags:
code-generation
License:
FudanSELab
commited on
Commit
•
772c997
1
Parent(s):
0933537
Update README.md
Browse files
README.md
CHANGED
@@ -37,21 +37,66 @@ For 412 methods, they have been constructed with diverse dependencies, including
|
|
37 |
|
38 |
### Languages
|
39 |
|
40 |
-
|
41 |
|
42 |
## Dataset Structure
|
43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
### Data Instances
|
45 |
|
46 |
[More Information Needed]
|
47 |
|
48 |
### Data Fields
|
49 |
|
50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
|
52 |
### Data Splits
|
53 |
|
54 |
-
|
55 |
|
56 |
## Dataset Creation
|
57 |
|
@@ -109,7 +154,16 @@ For 412 methods, they have been constructed with diverse dependencies, including
|
|
109 |
|
110 |
### Citation Information
|
111 |
|
112 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
|
114 |
### Contributions
|
115 |
|
|
|
37 |
|
38 |
### Languages
|
39 |
|
40 |
+
The programming language is Python. The natural language used in the comments and docstrings is English.
|
41 |
|
42 |
## Dataset Structure
|
43 |
|
44 |
+
```python
|
45 |
+
from datasets import load_dataset
|
46 |
+
dataset = load_dataset("FudanSELab/ClassEval")
|
47 |
+
|
48 |
+
DatasetDict({
|
49 |
+
test: Dataset({
|
50 |
+
features: ['task_id', 'skeleton', 'test', 'solution_code', 'import_statement', 'class_description', 'methods_info',
|
51 |
+
'class_name', 'test_classes', 'class_constructor', 'fields'],
|
52 |
+
num_rows: 100
|
53 |
+
})
|
54 |
+
})
|
55 |
+
```
|
56 |
+
|
57 |
### Data Instances
|
58 |
|
59 |
[More Information Needed]
|
60 |
|
61 |
### Data Fields
|
62 |
|
63 |
+
The specific data fields for each task are delineated as follows:
|
64 |
+
|
65 |
+
* task_id: the unique identifier for each task.
|
66 |
+
|
67 |
+
* skeleton: the class skeleton, including all input descriptions in our class-level coding tasks.
|
68 |
+
|
69 |
+
* test: all test cases for the whole class.
|
70 |
+
|
71 |
+
* solution_code: the ground-truth class-level code for each task.
|
72 |
+
|
73 |
+
More fine-grained class-level information from the class skeleton, including:
|
74 |
+
|
75 |
+
* import_statement: the import statements for each task.
|
76 |
+
|
77 |
+
* class_name: the name of the class.
|
78 |
+
|
79 |
+
* class_description: a concise description of the purpose and functionality of the class.
|
80 |
+
|
81 |
+
* class_constructor: the whole constructor of the class.
|
82 |
+
|
83 |
+
* fields: the fields defined in the class_constructor.
|
84 |
+
|
85 |
+
Detailed information for each method in the "methods_info" field, including:
|
86 |
+
|
87 |
+
* method_name: the method signature.
|
88 |
+
|
89 |
+
* method_input: the method contract design, including all input descriptions in the method.
|
90 |
+
|
91 |
+
* test_code: the test cases for the method.
|
92 |
+
|
93 |
+
* solution_code: the ground-truth method-level code.
|
94 |
+
|
95 |
+
* dependencies: the dependency information of the method.
|
96 |
|
97 |
### Data Splits
|
98 |
|
99 |
+
The dataset only consists of a test split with 100 samples.
|
100 |
|
101 |
## Dataset Creation
|
102 |
|
|
|
154 |
|
155 |
### Citation Information
|
156 |
|
157 |
+
```
|
158 |
+
@misc{du2023classeval,
|
159 |
+
title={ClassEval: A Manually-Crafted Benchmark for Evaluating LLMs on Class-level Code Generation},
|
160 |
+
author={Xueying Du and Mingwei Liu and Kaixin Wang and Hanlin Wang and Junwei Liu and Yixuan Chen and Jiayi Feng and Chaofeng Sha and Xin Peng and Yiling Lou},
|
161 |
+
year={2023},
|
162 |
+
eprint={2308.01861},
|
163 |
+
archivePrefix={arXiv},
|
164 |
+
primaryClass={cs.CL}
|
165 |
+
}
|
166 |
+
```
|
167 |
|
168 |
### Contributions
|
169 |
|