Datasets:
Size:
1M<n<10M
ArXiv:
Tags:
programming-language
code
program-synthesis
automatic-code-repair
code-retrieval
code-translation
License:
update data loader
Browse files- xCodeEval.py +32 -4
xCodeEval.py
CHANGED
@@ -148,8 +148,7 @@ _DESCRIPTIONS = {
|
|
148 |
18. `prob_desc_created_at`: The Unix timestamp when the problem was released. Use `datetime` lib in Python to parse it to a human-readable format.
|
149 |
19. `file_name`: Name of the source jsonl file from where data is loaded.
|
150 |
20. `hidden_unit_tests`: a list of unit tests returned as string. use `json.loads(hidden_unit_tests)` to load the data.
|
151 |
-
|
152 |
-
Objective: Given a source code in lang, generate a code in target lang."""
|
153 |
),
|
154 |
"program_synthesis": textwrap.dedent(
|
155 |
"""### Key Definitions
|
@@ -161,7 +160,20 @@ _DESCRIPTIONS = {
|
|
161 |
6. `code_uid`: A unique ID for the sample. It is not important for model training. If you find any issue with the sample, you can report it to us mentioning the `code_uid`.
|
162 |
7. `difficulty`: Difficulty rating of the problem indicated by `src_uid`. The higher the harder.
|
163 |
8. `exec_outcome`: Execution outcome status. Follow [Section 4.1](https://arxiv.org/pdf/2303.03004.pdf) to know the potential list of outcomes. The `exec_outcome` flags in the training data comes from a pre-run environmeent. However, training data doesn't includes unit-test to avoid potential hacks. We provide unit test for only dev and test data.
|
164 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
165 |
),
|
166 |
"retrieval_code_code": textwrap.dedent(
|
167 |
"""### Key Definitions
|
@@ -169,6 +181,7 @@ _DESCRIPTIONS = {
|
|
169 |
2. `negative_code` : list of negative codes for `nl`
|
170 |
3. `src_uid` : A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
171 |
4. `source_code` : A source code given as input query.
|
|
|
172 |
Objective: Given a source_code retrieve similar source code from `retrieval_corpus`."""
|
173 |
),
|
174 |
"retrieval_nl_code": textwrap.dedent(
|
@@ -177,12 +190,14 @@ _DESCRIPTIONS = {
|
|
177 |
2. `positive_code` : list of positive codes for `nl`
|
178 |
3. `negative_code` : list of negative codes for `nl`
|
179 |
4. `src_uid` : A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
|
|
180 |
Objective: Given a nl (problem description) retrieve similar source code from `retrieval_corpus`."""
|
181 |
),
|
182 |
"retrieval_corpus": textwrap.dedent(
|
183 |
"""### Key Definitions
|
184 |
1. `idx` : unique index for each sample on a specific langauge (read language from filename).
|
185 |
4. `source_code` : A source code given as retrieval document.
|
|
|
186 |
Objective: Use the retrival_corpus to perform query for retrieval_nl_code and retrieval_code_code ."""
|
187 |
),
|
188 |
"code_compilation": textwrap.dedent(
|
@@ -194,6 +209,7 @@ _DESCRIPTIONS = {
|
|
194 |
5. `code_uid`: A unique ID for the sample. It is not important for model training. If you find any issue with the sample, you can report it to us mentioning the `code_uid`.
|
195 |
6. `src_uid`: A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
196 |
7. `difficulty`: Difficulty rating of the problem indicated by `src_uid`. The higher the harder.
|
|
|
197 |
Objective: Given a `source_code` the objective is to classify if the code compiles or not (label:compilation_error) ."""
|
198 |
),
|
199 |
"tag_classification": textwrap.dedent(
|
@@ -205,7 +221,19 @@ _DESCRIPTIONS = {
|
|
205 |
5. `code_uid`: A unique ID for the sample. It is not important for model training. If you find any issue with the sample, you can report it to us mentioning the `code_uid`.
|
206 |
6. `src_uid`: A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
207 |
7. `difficulty`: Difficulty rating of the problem indicated by `src_uid`. The higher the harder.
|
208 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
209 |
),
|
210 |
}
|
211 |
|
|
|
148 |
18. `prob_desc_created_at`: The Unix timestamp when the problem was released. Use `datetime` lib in Python to parse it to a human-readable format.
|
149 |
19. `file_name`: Name of the source jsonl file from where data is loaded.
|
150 |
20. `hidden_unit_tests`: a list of unit tests returned as string. use `json.loads(hidden_unit_tests)` to load the data.
|
151 |
+
Objective: Given a source code (`source_code`) in `lang_cluster`, generate a code in target programming language."""
|
|
|
152 |
),
|
153 |
"program_synthesis": textwrap.dedent(
|
154 |
"""### Key Definitions
|
|
|
160 |
6. `code_uid`: A unique ID for the sample. It is not important for model training. If you find any issue with the sample, you can report it to us mentioning the `code_uid`.
|
161 |
7. `difficulty`: Difficulty rating of the problem indicated by `src_uid`. The higher the harder.
|
162 |
8. `exec_outcome`: Execution outcome status. Follow [Section 4.1](https://arxiv.org/pdf/2303.03004.pdf) to know the potential list of outcomes. The `exec_outcome` flags in the training data comes from a pre-run environmeent. However, training data doesn't includes unit-test to avoid potential hacks. We provide unit test for only dev and test data.
|
163 |
+
9. `prob_desc_description`: Problem description in textual format, math operations are written in latex.
|
164 |
+
10. `prob_desc_input_from`: How the program should take the unit test.
|
165 |
+
11. `prob_desc_output_to`: Where the program should output the result of the unit test.
|
166 |
+
12. `prob_desc_time_limit`: Time limit to solve the problem.
|
167 |
+
13. `prob_desc_memory_limit`: Memory limit to solve the problem.
|
168 |
+
14. `prob_desc_input_spec`: How and in what order the input will be given to the program? It also includes the date range, types, and sizes.
|
169 |
+
15. `prob_desc_output_spec`: How the outputs should be printed. Most of the time the unit test results are matched with an *exact string match* or *floating point comparison* with a precision boundary.
|
170 |
+
16. `prob_desc_sample_inputs`: A sample input for the code that is expected to solve the problem described in `description`.
|
171 |
+
17. `prob_desc_sample_outputs`: The expected output for the `sample_input` that is expected to solve the problem described in `description`.
|
172 |
+
18. `prob_desc_notes`: Explanation of `sample_inputs` & `sample_outputs`.
|
173 |
+
19. `prob_desc_created_at`: The Unix timestamp when the problem was released. Use `datetime` lib in Python to parse it to a human-readable format.
|
174 |
+
20. `file_name`: Name of the source jsonl file from where data is loaded.
|
175 |
+
21. `hidden_unit_tests`: a list of unit tests returned as string. use `json.loads(hidden_unit_tests)` to load the data.
|
176 |
+
Objective: Given a `src_uid` read problem description from `problem_descriptions.jsonl` and generate a solution for problem description."""
|
177 |
),
|
178 |
"retrieval_code_code": textwrap.dedent(
|
179 |
"""### Key Definitions
|
|
|
181 |
2. `negative_code` : list of negative codes for `nl`
|
182 |
3. `src_uid` : A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
183 |
4. `source_code` : A source code given as input query.
|
184 |
+
5. `file_name`: Name of the source jsonl file from where data is loaded.
|
185 |
Objective: Given a source_code retrieve similar source code from `retrieval_corpus`."""
|
186 |
),
|
187 |
"retrieval_nl_code": textwrap.dedent(
|
|
|
190 |
2. `positive_code` : list of positive codes for `nl`
|
191 |
3. `negative_code` : list of negative codes for `nl`
|
192 |
4. `src_uid` : A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
193 |
+
5. `file_name`: Name of the source jsonl file from where data is loaded.
|
194 |
Objective: Given a nl (problem description) retrieve similar source code from `retrieval_corpus`."""
|
195 |
),
|
196 |
"retrieval_corpus": textwrap.dedent(
|
197 |
"""### Key Definitions
|
198 |
1. `idx` : unique index for each sample on a specific langauge (read language from filename).
|
199 |
4. `source_code` : A source code given as retrieval document.
|
200 |
+
3. `file_name`: Name of the source jsonl file from where data is loaded.
|
201 |
Objective: Use the retrival_corpus to perform query for retrieval_nl_code and retrieval_code_code ."""
|
202 |
),
|
203 |
"code_compilation": textwrap.dedent(
|
|
|
209 |
5. `code_uid`: A unique ID for the sample. It is not important for model training. If you find any issue with the sample, you can report it to us mentioning the `code_uid`.
|
210 |
6. `src_uid`: A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
211 |
7. `difficulty`: Difficulty rating of the problem indicated by `src_uid`. The higher the harder.
|
212 |
+
8. `file_name`: Name of the source jsonl file from where data is loaded.
|
213 |
Objective: Given a `source_code` the objective is to classify if the code compiles or not (label:compilation_error) ."""
|
214 |
),
|
215 |
"tag_classification": textwrap.dedent(
|
|
|
221 |
5. `code_uid`: A unique ID for the sample. It is not important for model training. If you find any issue with the sample, you can report it to us mentioning the `code_uid`.
|
222 |
6. `src_uid`: A specific identifier that shows which problem the code is associated with. This identifier is **important** for the training of the model. The problem referred to by the `src_uid` provides a natural description of the problem that the code successfully solved. Refer to [Structure of `problem_descriptions.jsonl`](./README.md#structure-of-problem_descriptionsjsonl)
|
223 |
7. `difficulty`: Difficulty rating of the problem indicated by `src_uid`. The higher the harder.
|
224 |
+
8. `prob_desc_description`: Problem description in textual format, math operations are written in latex.
|
225 |
+
9. `prob_desc_input_from`: How the program should take the unit test.
|
226 |
+
10. `prob_desc_output_to`: Where the program should output the result of the unit test.
|
227 |
+
11. `prob_desc_time_limit`: Time limit to solve the problem.
|
228 |
+
12. `prob_desc_memory_limit`: Memory limit to solve the problem.
|
229 |
+
13. `prob_desc_input_spec`: How and in what order the input will be given to the program? It also includes the date range, types, and sizes.
|
230 |
+
14. `prob_desc_output_spec`: How the outputs should be printed. Most of the time the unit test results are matched with an *exact string match* or *floating point comparison* with a precision boundary.
|
231 |
+
15. `prob_desc_sample_inputs`: A sample input for the code that is expected to solve the problem described in `description`.
|
232 |
+
16. `prob_desc_sample_outputs`: The expected output for the `sample_input` that is expected to solve the problem described in `description`.
|
233 |
+
17. `prob_desc_notes`: Explanation of `sample_inputs` & `sample_outputs`.
|
234 |
+
18. `prob_desc_created_at`: The Unix timestamp when the problem was released. Use `datetime` lib in Python to parse it to a human-readable format.
|
235 |
+
19. `file_name`: Name of the source jsonl file from where data is loaded.
|
236 |
+
Objective: Given a `source_code` the objective is to classify the code into multi-label tags (label:tags)."""
|
237 |
),
|
238 |
}
|
239 |
|