feat (search): adding option to disable dumping the results to civ #325
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #325 +/- ##
===========================================
+ Coverage 43.69% 49.92% +6.23%
===========================================
Files 124 108 -16
Lines 8377 7609 -768
Branches 1375 1219 -156
===========================================
+ Hits 3660 3799 +139
+ Misses 4387 3468 -919
- Partials 330 342 +12 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
| if len(resultsList) > 0: | ||
| started_dumping = self._csv_cursor > 0 | ||
| file_mode = "a" if started_dumping else "w" | ||
|
|
||
| if not (started_dumping): | ||
| for result in resultsList: | ||
| # Waiting to start receiving non-failed jobs before dumping results | ||
| is_single_obj_and_has_success = ( | ||
| "objective" in result and type(result["objective"]) is not str | ||
| ) | ||
| is_multi_obj_and_has_success = ( | ||
| "objective_0" in result and type(result["objective_0"]) is not str | ||
| ) | ||
| if is_single_obj_and_has_success or is_multi_obj_and_has_success or flush: | ||
| self._csv_columns = result.keys() | ||
| break | ||
|
|
||
| if self._csv_columns is not None: | ||
| with open(os.path.join(path), file_mode) as fp: | ||
| writer = csv.DictWriter(fp, self._csv_columns, extrasaction="ignore") | ||
| if not (started_dumping): | ||
| writer.writeheader() | ||
| writer.writerows(resultsList) | ||
| self._csv_cursor += len(resultsList) |
There was a problem hiding this comment.
Can this be avoided by converting the results list to a dataframe then use pandas to write the dataframe to csv file? This would eliminate importing the csv module.
There was a problem hiding this comment.
This logic is to avoid redundant writing. It just appends new lines. The pandas logic would re-write everything again an again. Making the complexity squared instead of linear.
There was a problem hiding this comment.
The self.jobs = [] is a list of dictionaries, is that right? Each dictionary represents the results of a completed job, is that correct?
There was a problem hiding this comment.
The self.jobs is actually a list of HPOJob objects.
|
At the end of the self.dump_jobs_done_to_csv(flush=True)
self.history.compute_pareto_efficiency()
df_results = self.history.to_dataframe()
return df_resultsIt looks like the job results are written to CSV then the pareto results are calculated. Does this update the CSV with the pareto results? |
…gather callback for callback interface, cleaning up evaluator code, adapting code in search
|
The PR is looking better but we still need to add a few unit tests. |
|
The |
|
I merged the latest develop branch into this branch and apparently it fixed the failing Redis tests. |
|
For the |
|
The search method is currently defined as: search(self, max_evals: int = -1, timeout: int = None, max_evals_strict: bool = False)A better definition would be the following: search(self, max_evals=-1, timeout=0, max_evals_strict=False) -> pd.DataFrameWhen you have default values there is no need to explicitly give the type, it is inferred from the default value. I don't know why |
|
Hi @wigging ,
|
Deathn0t
changed the title
Allow the user to disable writing search results to a CSV file. Collect search results using a SearchHistory class. Resolves issue #309 .