weixin_39693971的博客

Hello, Expressions as follows : ^(.{1,35}(\n|$)){0,4}$ does does not appear to be working properly. Even with the last update, the dot '.' with the alternators '()' is no longer supported. My thanks in advance for your attention.

• Thanks for the quick response.

What sort of tweaks are needed to vendor/autoload.php?

It's not solved, so it should not be closed.

But judging from the screenshot above, this would be correct.

Ignoring the overhead of loading one additional unused datapoint a simple fix for both (relative and absolue windowSize) could be:

diff

-    for i in range(windowPoints, len(series)):
+    for i in range(windowPoints + 1, len(series)):
      nonNull = [v for v in series[i - windowPoints:i] if v is not None]

I guess the error is at https://github.com/graphite-project/graphite-web/blob/17a34e7966f7a46eded30c2362765c74eea899cb/webapp/graphite/render/functions.py#L1136-L1137

For a windowSize of 1 series[i - windowPoints:i] is actually using the previous datapoint series[1 - 1:1] which is series[0:1] including series[0] but excluding series[1] to calculate the first datapoint for the result. Since series includes one additional datapoint from the past, the past datapoint is used to calculate the first resulting datapoint.

Hi must forget to publish my comment but

what I was writing was:

I humbly think it better to make a branch for bootstrap4 than when it bootstrap3 become expire or boostrap4 is bulletproof ready for migration merge it with the master branch

just to mention the cookiesandyou is already included in bootstrap4-blog theme https://github.com/alanorth/hugo-theme-bootstrap4-blog/blob/master/exampleSite/config.toml#L56

this might help: https://agrimprasad.com/post/hugo-goldmark-markdown/#render-hooks-with-goldmark

Must call Vitamio.isInitialized(this); after super.onCreate(); but not onStart(); Must call Vitamio.isInitialized(this); after super.onCreate(); but not onStart();

my PS make me a fool because you ask for Minimum Hugo Version: 0.30

Hi again;

I reopen this issue because I have it again I try to install our Install Academic with Netlify which works fine but after when I clone my github repo and try to execute hugo, I have all these error :

Started building sites ...
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/about.md: (4, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/posts.md: (5, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/projects.md: (4, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/publications.md: (5, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/contact.md: (4, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/publications_selected.md: (6, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/tags.md: (4, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/talks.md: (4, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for post/_index.md: (2, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for home/teaching.md: (6, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for talk/_index.md: (2, 12): unexpected token
ERROR 2018/05/05 07:06:53 failed to parse page metadata for publication/_index.md: (2, 12): unexpected token
Error: Error building site: Errors reading pages: Error: failed to parse page metadata for home/about.md: (4, 12): unexpected token for about.md
Error: failed to parse page metadata for home/posts.md: (5, 12): unexpected token for posts.md
Error: failed to parse page metadata for home/projects.md: (4, 12): unexpected token for projects.md
Error: failed to parse page metadata for home/publications.md: (5, 12): unexpected token for publications.md
Error: failed to parse page metadata for home/contact.md: (4, 12): unexpected token for contact.md
Error: failed to parse page metadata for home/publications_selected.md: (6, 12): unexpected token for publications_selected.md
Error: failed to parse page metadata for home/tags.md: (4, 12): unexpected token for tags.md
Error: failed to parse page metadata for home/talks.md: (4, 12): unexpected token for talks.md
Error: failed to parse page metadata for post/_index.md: (2, 12): unexpected token for _index.md
Error: failed to parse page metadata for home/teaching.md: (6, 12): unexpected token for teaching.md
Error: failed to parse page metadata for talk/_index.md: (2, 12): unexpected token for _index.md
Error: failed to parse page metadata for publication/_index.md: (2, 12): unexpected token for _index.md

Also If I try to import it on forestry It works partially with this error :

Failed to parse file: themes/academic/archetypes/talk.md
Failed to parse input on line 1 at offset 9
date =   # Schedule page publish date.

         ^

How could I correct this ?

Regards!

Jonathan

PS I use : Hugo Static Site Generator v0.20.2 darwin/amd64 BuildDate: 2018-05-04T15:29:07-04:00

I realy try to play the dumb here and still not working

git clone https://github.com/sourcethemes/academic-kickstart.git My_Website
Cloning into 'My_Website'...
remote: Counting objects: 77, done.
remote: Total 77 (delta 0), reused 0 (delta 0), pack-reused 77
Unpacking objects: 100% (77/77), done.

cd My_Website/

hugo serve
Started building sites ...
ERROR 2018/04/09 02:50:43 failed to parse page metadata for "home/publications.md": Near line 5, key 'date': Near line 6: Expected 'T' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:50:43 failed to parse page metadata for "home/publications_selected.md": Near line 6, key 'date': Near line 7: Expected 'T' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:50:43 failed to parse page metadata for "home/tags.md": Near line 4, key 'date': Near line 5: Expected 'T' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:50:43 failed to parse page metadata for "post/_index.md": Near line 2, key 'date': Near line 3: Expected 'T' in ISO8601 datetime, but found '\n' instead.
Error: Error building site: Errors reading pages: Error: failed to parse page metadata for "home/publications.md": Near line 5, key 'date': Near line 6: Expected 'T' in ISO8601 datetime, but found '\n' instead. for publications.md
Error: failed to parse page metadata for "home/publications_selected.md": Near line 6, key 'date': Near line 7: Expected 'T' in ISO8601 datetime, but found '\n' instead. for publications_selected.md
Error: failed to parse page metadata for "home/tags.md": Near line 4, key 'date': Near line 5: Expected 'T' in ISO8601 datetime, but found '\n' instead. for tags.md
Error: failed to parse page metadata for "post/_index.md": Near line 2, key 'date': Near line 3: Expected 'T' in ISO8601 datetime, but found '\n' instead. for _index.md


git submodule update --init --recursive
Submodule 'themes/academic' (https://github.com/gcushen/hugo-academic.git) registered for path 'themes/academic'
Cloning into '/home/me/Public/hugo/My_Website/themes/academic'...
Submodule path 'themes/academic': checked out 'b02f5ecedabe4494dabee522db3ab843472afa41'
[me My_Website]$ hugo serve
Started building sites ...
ERROR 2018/04/09 02:51:05 failed to parse page metadata for "home/publications.md": Near line 5, key 'date': Near line 6: Expected 'T' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:51:05 failed to parse page metadata for "home/tags.md": Near line 4, key 'date': Near line 5: Expected 'T' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:51:05 failed to parse page metadata for "post/_index.md": Near line 2, key 'date': Near line 3: Expected 'T' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:51:05 failed to parse page metadata for "home/publications_selected.md": Near line 6, key 'date': Near line 7: Expected 'T' in ISO8601 datetime, but found '\n' instead.
Error: Error building site: Errors reading pages: Error: failed to parse page metadata for "home/publications.md": Near line 5, key 'date': Near line 6: Expected 'T' in ISO8601 datetime, but found '\n' instead. for publications.md
Error: failed to parse page metadata for "home/tags.md": Near line 4, key 'date': Near line 5: Expected 'T' in ISO8601 datetime, but found '\n' instead. for tags.md
Error: failed to parse page metadata for "post/_index.md": Near line 2, key 'date': Near line 3: Expected 'T' in ISO8601 datetime, but found '\n' instead. for _index.md
Error: failed to parse page metadata for "home/publications_selected.md": Near line 6, key 'date': Near line 7: Expected 'T' in ISO8601 datetime, but found '\n' instead. for publications_selected.md

sorry I gave up even after this it result to a blank page

than even fixing this one too your theme restart giving me error like this

Started building sites ...
ERROR 2018/04/09 02:42:16 failed to parse page metadata for "home/hero.md": Near line 23, key 'cta.label': Near line 23: Expected value but found '\'' instead.
ERROR 2018/04/09 02:42:16 failed to parse page metadata for "post/getting-started.md": Near line 4, key 'lastmod': Near line 5: Expected 'T' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:42:16 failed to parse page metadata for "talk/example-talk.md": Near line 4, key 'time_start': Near line 5: Expected 'Z' in ISO8601 datetime, but found '\n' instead.
ERROR 2018/04/09 02:42:16 failed to parse page metadata for "publication/clothing-search.md": Near line 50, key 'url_custom': Near line 50: Expected value but found '{' instead.
Error: Error building site: Errors reading pages: Error: failed to parse page metadata for "home/hero.md": Near line 23, key 'cta.label': Near line 23: Expected value but found '\'' instead. for hero.md
Error: failed to parse page metadata for "post/getting-started.md": Near line 4, key 'lastmod': Near line 5: Expected 'T' in ISO8601 datetime, but found '\n' instead. for getting-started.md
Error: failed to parse page metadata for "talk/example-talk.md": Near line 4, key 'time_start': Near line 5: Expected 'Z' in ISO8601 datetime, but found '\n' instead. for example-talk.md
Error: failed to parse page metadata for "publication/clothing-search.md": Near line 50, key 'url_custom': Near line 50: Expected value but found '{' instead. for clothing-search.md

So I would understood if you said you team was in dev but as far I look you give the impression everything is solid and well done, which, at least for my case is not.

at https://sourcethemes.com/academic/docs/offline-site/ it could be interesting to define what means

Add Font Awesome font assets to static/

since to make it work I had to copy webfonts in static/css/ instead of static because with https://cdnjs.cloudflare.com/ajax/libs/font-awesome/5.11.2/css/all.min.css point to ../webfonts and not ../../webfonts

said, too many project are US Centric, such as S3, which is the backend of Github is based in US, often during COVID I had to use a VPN to being directly connected in US to be able to download and clone project which relay on github and S3. (I'm based in Bangkok which is SEA) without this VPN I always had timeout and/or downloading a 3 bites second.

Also by using "offline site" version of my site, I speed up by 200% the rendering and now this one could be cached entirely by CloudFlare which means my website is now available in 200 cities.

result with offline site options enabled

on netlify

on cloudflare as workers

result without offline site

such as the update_academic.sh script, I think it could be easy to make a script to parse and download all these resources, of course I'm not a Windows user.

but personally I simply don't understand having everything locally is not the default behavior instead of relaying on a bunch of uncacheable externals ressources.

If other show interest, yes it would be a nice addon

however I'm happy with this workaround

`
+++
widget = "slider"
weight = 11

interval = 4200
height = "25vh"

[design.background]
  gradient_start = "#0000F9"
  gradient_end = "#86007D"
  text_color_light = true

[[item]]
  title = "Center"
  content = "I am center aligned :smile:"
  cta_label = "Get Academic"
  cta_url = "https://sourcethemes.com/academic/"
  cta_icon_pack = "fas"
  cta_icon = "graduation-cap"

[[item]]
  title = "Left"
  content = "I am left aligned :smile:"

[[item]]
  title = "Right"
  content = "I am right aligned :smile:"
+++

`

I had the same question

what resolution (I mean size and ratio in pixels) is recommended for these background images? in fact in _widgets.scss under .wg-hero the background-size is cover

JFYI I don't know what happened between March 2 (the day I opened this issue) and March 7 (the day you release version 4.8) but after losing few days of debugging my pseudo error which you believe it was on my side and too complex to support it has been fixed automagically with your "official" release

thanks for your reply and sorry I missed the template For **help** and **questions**, please join our **[community chat](https://spectrum.chat/academic)** or use the **[forum](https://discourse.gohugo.io/c/themes)** 🚑.

I might skip a version I think my bug is related to this refactor in v4.7 - Refactor cookie consent so that it will use local assets for offline sites

yes, it should be a new version that is backward incompatible.

when he has time. In the meantime you can use my fork to make it work with 5.8, maybe you find other errors to be fixed.

Add this to your composer.json:

json
"repositories": [
    {
      "type": "vcs",
      "url": "https://github.com/graphicms/laravel-mongodb"
    }
  ]

It does not seem like it does :) I use my fork now in 5.8 until this will be usable in 5.8 and everything works.

https://github.com/laravel/framework/pull/26391

got it :) I created the branch to check if it works locally with L7 and it did :)

Hmm guess we have a breaking change:

PHP Fatal error:  Access level to Jenssegers\Mongodb\Relations\BelongsToMany::newPivotQuery() must be public (as in class Illuminate\Database\Eloquent\Relations\BelongsToMany) in /home/runner/work/laravel-mongodb/laravel-mongodb/src/Jenssegers/Mongodb/Relations/BelongsToMany.php on line 268

强烈支持

I have pretty much the same problem. And I may have a theory where this comes from. I have a list of files I render in a sidebar. Everything is perfect, but obviously the user can change the name of these files. The files have an ID, which in my case is a simple hash of the absolute file path. So on rename, this file path changes, and so does the hash a.k.a. the ID.

If then I re-order the items (because one can sort the directory differently, obviously), the bug appears with the renamed file actually appearing twice, even though the hash changed, which means that the respective component is being re-used and then rendered on the original position (not the new one in the different sorting).

Here's how it looks:

As you can see, the file list itself only contains the file once, but it is being rendered twice. Completely removing the fileList and using a timeout to re-fill the list after a DOM rendering reflow fixes this, but obviously with a display glitch as the empty list is being rendered for a fraction of a second.

So it seems that the bug originates in the components not being re-validated. It doesn't check whether or not the respective hash/keyField exists more than once.

EDIT: I should add that another file is below the duplicate renamed file, so the duplicate file actually perfectly covers the file that's supposed to show up at the old position.

almost the same for me. looks like it fails with activities extend FragmentActivity from support library

java.lang.ClassCastException: java.lang.NoClassDefFoundError cannot be cast to java.lang.RuntimeException
    at org.robolectric.internal.ReflectionHelpers.callInstanceMethodReflectively(ReflectionHelpers.java:68)
    at org.robolectric.util.ActivityController$1.run(ActivityController.java:115)
    at org.robolectric.shadows.ShadowLooper.runPaused(ShadowLooper.java:268)
    at org.robolectric.util.ActivityController.create(ActivityController.java:111)
    at org.robolectric.util.ActivityController.create(ActivityController.java:122)

can't wait for the fix) thanks

Pull request: https://github.com/Bttstrp/bootstrap-switch/pull/680

Still waiting after 10 days...

Is anyone managing this repository ?

I just sent a pull request to do so, waiting for approval.

vivo手机遇到了同样的问题，这个怎么解

Can also confirm that this works for me. However, when tabs are activated, empty lines are added for every entry in the nav section of the mkdocs.yml. Has anybody else experienced that?

Same issue here, I found that adding subnets to the endpoint doesn't work, you need to create it with the subnets already defined. This is true also if adding subnets to the already created endpoint in the AWS Console, it throws an error when applying.

's solution works when creating from scratch, the only difference is that I'm not creating the list from a data resource:

yaml
resource "aws_vpc_endpoint" "ec2" {
  vpc_id              = "${aws_vpc.main.id}"
  service_name        = "com.amazonaws.${var.region}.ec2"
  vpc_endpoint_type   = "Interface"
  security_group_ids  = ["${aws_security_group.private.id}"]
  private_dns_enabled = true
  subnet_ids          = ["${formatlist("%s", aws_subnet.private.*.id)}"]
}

, this is ready for review

Last one. test/test_metalearning/test_metalearning.py uses some constants like ACC_METRIC and AUC_METRIC (line 36 onwards). I couldn't find the definition of those as well.

test/test_pipeline/components/dummy_components/dummy_component_import.py is 100% PEP8 deviating. :-D Is this file still necessary?

Yes. Please put a # noqa there to ignore the PEP8 error(s).

ok

test_metalearning.py uses two functions (calc_meta_features and calc_meta_features_encoded ) whose definitions I couldn't find. I think this didn't cause errors before due to star imports. What should we do here?

Apparently, that unit test was skipped. You can find those functions in autosklearn.smbo

Ok. They have slightly different names over there. That's why I couldn't find them. For instance _calculate_metafeatures instead of calc_meta_features.

test/test_pipeline/components/dummy_components/dummy_component_import.py is 100% PEP8 deviating. :-D Is this file still necessary?

test_metalearning.py uses two functions (calc_meta_features and calc_meta_features_encoded ) whose definitions I couldn't find. I think this didn't cause errors before due to star imports. What should we do here?

I have exactly this same issue. Disabling Zen Coding fixes it; but that creates another problem: no zen coding. Obviously this isn't the zen coding board; but does anyone know if there is a way to disable zen coding for stylus files?

I'm having the same issue with IAM certificates on Cloudfront.

In the IAM resource I have:

tf
lifecycle {
    create_before_destroy = true
  }

which allows me to create a new certificate and attach it to Cloudfront, but then terraform tries to delete the certificate and it fails, because Cloudfront takes as long as 15m to completely apply changes. The error message is:

tf
* module.group4_cdn_certificate.aws_iam_server_certificate.iam_server_certificate (destroy): 1 error(s) occurred:

* aws_iam_server_certificate.iam_server_certificate (deposed #0): 1 error(s) occurred:

* aws_iam_server_certificate.iam_server_certificate (deposed #0): DeleteConflict: Certificate: XXXXXXXXXXXXXXXXXXXXX is currently in use by CloudFront Distribution XXXXXXXXXXXX. Please remove it first before deleting it from IAM.
    status code: 409, request id: xxxxxxxx-xxxx-11e8-bb86-df946c6766f3

I've tried to use a local-exec provisioner:

tf
provisioner "local-exec" {
    when    = "destroy"
    command = "sleep 900"
  }

but there is currently a bug, where the local-exec is not executed when you have create_before_destroy configured (issue -> https://github.com/hashicorp/terraform/issues/13549).

My current "solution" is allowing the apply to fail and then re-running plan & apply after 15m to just remove the certificate.

, should we close this one?

, thanks for the easy to understand example and for bringing up these issues. First, about the set/subset of configurations, I tried to reproduce this weird behavior, but couldn't. I tested many seeds and always got the expected answer:

Number of evaluated configurations
for model1 =  X
for model2 =  Y
for model3 =  Z

Number of common configurations
model 1 and model 2 = X
model 1 and model 3 = X
model 2 and model 3 = Y

I ran your example with auto-sklearn's latest version of the development branch, which is pretty much the next release that should be out in a couple of days. So what I recommend you to do is to updated your auto-sklearn (as soon as the new version sees the light of day) and rerun your code. If the issue persists, we would be glad to take a look at it again.

About the mean_fit_time, this is a SMAC thing. You can avoid this problem by specifying smac_scenario_args = {'initial_incumbent': 'RANDOM'} in the constructor of AutoSklearnClassifier.

To sum up, here is the code I ran and the output I got:

python
from sklearn.datasets import load_digits
from sklearn.model_selection import train_test_split
from autosklearn.classification import AutoSklearnClassifier

total_time = 50
time_per_task = 25
memory_limit= 32000  #does not work on jupyter lab for smaller values?

X, y = load_digits(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=123)

tmp_folder='./tmp1'
output_folder='./out1' 

seed = 1

automl1 = AutoSklearnClassifier( 
    time_left_for_this_task=total_time,
    ml_memory_limit= memory_limit,  
    per_run_time_limit=time_per_task,
    tmp_folder=tmp_folder, 
    output_folder=output_folder, 
    seed=seed, 
    shared_mode=True, 
    ensemble_size=0,
    initial_configurations_via_metalearning=0,
    )
automl1.fit(X_train, y_train)

seed = 2
automl2 = AutoSklearnClassifier( 
    time_left_for_this_task=total_time,
    ml_memory_limit= memory_limit,
    per_run_time_limit=time_per_task,
    tmp_folder=tmp_folder, 
    output_folder=output_folder, 
    seed=seed, 
    shared_mode=True, 
    ensemble_size=0,
    initial_configurations_via_metalearning=0,
    smac_scenario_args = {'initial_incumbent': 'RANDOM'}  # Added argument
    )
automl2.fit(X_train, y_train)

seed = 3
automl3 = AutoSklearnClassifier( 
    time_left_for_this_task=total_time,
    ml_memory_limit= memory_limit,
    per_run_time_limit=time_per_task,
    tmp_folder=tmp_folder, 
    output_folder=output_folder, 
    seed=seed, 
    shared_mode=True, 
    ensemble_size=0,
    initial_configurations_via_metalearning=0,
    smac_scenario_args = {'initial_incumbent': 'RANDOM'}  # Added argument
    )
automl3.fit(X_train, y_train)

# Combine the results from the 3 runs
seed = 4
ensemble = AutoSklearnClassifier( 
    time_left_for_this_task=total_time,
    ml_memory_limit= memory_limit,
    per_run_time_limit=time_per_task,
    tmp_folder=tmp_folder, 
    output_folder=output_folder, 
    seed=seed, 
    shared_mode=True, 
    ensemble_size=1,
    initial_configurations_via_metalearning=0
    )

ensemble.fit(X_train, y_train)


print("Number of evaluated configurations")
print("for model1 = ",len(automl1.cv_results_['mean_test_score']))
print("for model2 = ",len(automl2.cv_results_['mean_test_score']))
print("for model3 = ",len(automl3.cv_results_['mean_test_score']))
print("for ensemble = ",len(ensemble.cv_results_['mean_test_score']))

print("Number of common configurations")
print("model 1 and model 2 =",len([x for x in automl1.cv_results_['mean_test_score'] if x in automl2.cv_results_['mean_test_score']]))
print("model 1 and model 3 =",len([x for x in automl1.cv_results_['mean_test_score'] if x in automl3.cv_results_['mean_test_score']]))
print("model 2 and model 3 =",len([x for x in automl2.cv_results_['mean_test_score'] if x in automl3.cv_results_['mean_test_score']]))

print("Model mean fit times")
print("model1 = ",automl1.cv_results_['mean_fit_time'])
print("model2 = ",automl2.cv_results_['mean_fit_time'])

OUTPUT: Number of evaluated configurations for model1 = 6 for model2 = 13 for model3 = 19 for ensemble = 22 Number of common configurations model 1 and model 2 = 6 model 1 and model 3 = 6 model 2 and model 3 = 13 Model mean fit times model1 = [ 0.85856581 1.08830547 3.45141554 0.93370342 25.0317173 1.94852448] model2 = [ 0.65213227 0.85856581 1.08830547 3.45141554 0.93370342 25.0317173 1.94852448 0.64790392 25.02988791 1.08440471 1.4163847 0.62608266 1.77576828]

Dear , I will start working on this issue now. But before I do that, since you open it six months ago, I just wanted to ask you if you have gained any more insights into this problem since then. Have you noticed any other strange/interesting behavior related to this issue?

A template system should not be opinionated. If you need to a dynamic language for constructing dynamic messages, async requirement is legitimate. Why close this?

as of v0.11.279 the mono build & standalone build doesnt startup web UI blank & in log it says 2019-05-05 02:08:50.0596 Info Starting Jackett v0.11.279.0 - it hangs here & does nothing even after i delete all config the default port does not work either...

there are no console/network errors in inspector its like jackett is not even running.

EDIT: was able to fix by removing jackett completely from system & mono + config file then system reboot & the fresh install started working again.. then restored config files & all was good. + installed fresh mono-mdk

Ok, I tested all the combinations between n_jobs, temp_folder and delete_tmp_folder_after_terminate. And here are the results

| n_jobs | tmp_folder | delete_tmp_folder_after_terminate | Deletes? | |--------|------------|-------------------|---------| | None | None | False | False | | None | None | True | True | | None | specified | False | False | | None | specified | True | True | | 2 | None | False | False | | 2 | None | True | False | | 2 | specified | False | False | | 2 | specified | True | False |

As you can see, the behavior is actually independent of temp_folder. So I suggest the following docstring:

Removes tmp_folder when finished. However, if auto-sklearn is run in parallel, tmp_folder won't be deleted, even if delete_tmp_folder_after_terminate is True.

What do you think?

, I ran a test with the following parameters:

n_jobs=None, tmp_folder=None, output_folder=None, delete_tmp_folder_after_terminate=False, delete_output_folder_after_terminate=False

The old docstring says: "If tmp_folder is None tmp_dir will always be deleted". So, if the docstring is correct, we would expect that the folders would be deleted, even though the delete_after_terminate parameters are set to False. Right?

But that is not the case. The folders stay there. So unless I'm missing something here, I think the old docstring was already inaccurate. Then maybe we could leave the new one as short as "remove tmp_folder, when finished. Currently, tmp_folder won't be deleted if n_jobs is different than 1 or None"

Does all this make sense?

Thanks, this is a helpful improvement of the docs. How about being a bit more specific and changing the last sentence to: "If not running in parallel and tmp_folder is None, then tmp_dir will always be deleted unless this argument is False."?

Your are right, I just tested it and this last bit is not right. However, I just kind of used what was already there before: "If tmp_folder is None tmp_dir will always be deleted". I now wonder if this sentence was indeed correct.

A possible fix has just been merged into the master of scikit-learn. Therefore, I suggest closing this issue for now. https://github.com/scikit-learn/scikit-learn/pull/16718

Hi , thanks for investigating this issue. By the way, I think this problem is an old friend auto-sklearn. I call it the "np.newaxis thing". But I never bother investigating it because, as you pointed out, the error occurs stochastically. Usually we "get around it" just by re-running the test. But today I looked at it more carefully and I think I understand now what is causing the problem.

I suspect the problem is on scikit-learn, more preciselly in these lines of

_kernel_pca.py

python
        # flip eigenvectors' sign to enforce deterministic output
        self.alphas_, _ = svd_flip(self.alphas_,
                                   np.empty_like(self.alphas_).T)

https://github.com/scikit-learn/scikit-learn/blob/b853653234b6b0c15b2d2501db2e72041005f813/sklearn/decomposition/_kernel_pca.py#L217-L219

TLDR: Replace

np.empty_like()

np.ones_like()

Why?

The function

svd_flip

self.alphas_ 

np.empty_like()

_ 

(Here is the code for

svd_flip

Ok, but the problem is that

np.empty()

np.empty()

svd_flip

python
v *= signs[:, np.newaxis]

https://github.com/scikit-learn/scikit-learn/blob/b853653234b6b0c15b2d2501db2e72041005f813/sklearn/utils/extmath.py#L528

So, to fix this I substituted

np.empty_like()

np.zeros_like()

np.ones_like()

, It would be great if you could try this fix as well and give us a feedback.

Cheers