Current front-end does not have this. Would be good to have a feature to reset the run
https://www.openml.org/api_docs#!/run/get_run_reset_id

From @joaquinvanschoren on February 26, 2017 16:54

For any dataset that's yours, click edit. The form asks for the dataset again although this should not be possible.

Also check that editing becomes impossible when the dataset is active.

Copied from original issue: #95

From @amueller on October 13, 2017 9:7

I don't seem to be able to search by multiple tag via the web interface. I'm not sure if it's a restriction of the web interface or the server.
I want to look for datasets that are tagged as openml100 and trivial. There is at least one: https://www.openml.org/d/1466

Copied from original issue: #200

We need to add a notice to be GDPR compliant. Here is a generator:
https://www.freeprivacypolicy.com/index-gdpr.php?utm_expid=87121688-128.1rKgFXB3TSykKooC7u4siw.1&utm_referrer=https%3A%2F%2Fwww.google.de%2F

Example: https://www.openml.org/r/8867786
This happens when the task is supervised but the underlying dataset does not have a default target set.

From @HeidiSeibold on August 23, 2016 9:51

When I want to download a data set to R, I need the ID
getOMLDataSet(did = 213L)

It is hard to find when I look at the website. I could only find it in the link.

Copied from original issue: #56

From @joaquinvanschoren on September 19, 2017 15:15

Add a button to the study page that downloads the dataset list, meta-features, run details and evaluation results as a zip file.

Copied from original issue: #182

So I tried logging in with my old OpenML account and it displays an error that my account is inactive but does not provide any further options. What is the procedure in this case if there is any of course?

I just wrote the functionality. Will be documented soon. Similar to #4 but for datasets

From @joaquinvanschoren on April 27, 2016 9:33

When you create a task through the website and forget something, the warning is not shown. The form just reloads without further information.

Copied from original issue: #43

Looking at the website it's impossible to see what the validation strategy for a given task is, from what I can tell.

(previously erroneously posted here openml/openml-python#513)

From @mfeurer on May 23, 2017 14:2

Copied from original issue: #138

cloning a repo called "website" is really not that helpful ;)

From @amueller on July 26, 2017 19:32

Is there a way to search datasets by feature type via the website?
If so, it's not clear to me. It's also not entirely clear to me what possible feature types are.
From here it looks like there are "numeric", "nominal", "string" and "unknown". Is that documented somewhere?

Copied from original issue: #164

From @joaquinvanschoren on January 25, 2017 14:22

Datasets can have missing target values. It should not be possible to define a supervised task on them.

Copied from original issue: #84

I'm trying to look at a flow with many runs:
https://www.openml.org/f/6952

and it loads for a long time and kinda kills my browser. I assume it tries to request all runs? Maybe we can do something smarter?

Preparing for the discussion tomorrow, I'm trying to upload this file: https://github.com/ramp-kits/mouse_cytometry/blob/master/data/train.arff. I get an error with the attached screen. Not sure what the problem is. I created the arff by hand from a csv, maybe there is a syntax error, but I'm not sure how to debug this or correct the file.

From @amueller on October 30, 2017 16:25

I just found out that you could click on "response schema" in the REST docs to see the schema. Even though it's not always useful, at least it's a bit useful. The styling didn't really convey to me that this is an expandable section. Maybe add a downward triangle next to it and an upward triangle next to Response sample to indicate that section is expanded?

Copied from original issue: #202

Right now all the pages on openml.org have title "OpenML". So in the bookmarks and browser history they all look the same making navigation difficult.

From @amueller on November 15, 2017 20:24

If you go to https://www.openml.org/api_docs#!/data/get_data_list
and click on the get button or the text next to it, you are redirected to https://www.openml.org/#!/data/get_data_list which should probably be https://www.openml.org/api_docs#!/data/get_data_list instead (it takes you to the landing page)

Copied from original issue: #208

From @amueller on March 14, 2016 11:44

When using a password <8 characters, the form still allows me to submit, but then fails, and resets all the fields. I should not be able to submit without having matching passwords of the right length.

Copied from original issue: #39

From @HeidiSeibold on October 12, 2017 12:47

When I look at e.g. the mean absolute error
https://www.openml.org/a/evaluation-measures/mean-absolute-error
it is not super useful.

I think we should

• clean it up with
  • no unrendered formulas
  • correct specification of the minimum and maximum
  • a working link to the code
• add information on which task type this can/should be used for.

Copied from original issue: #197

From @gabartas on October 12, 2017 12:48

Hello!
While navigating the website, I noticed that if the page started to render a chart or table (like in the image), clicking on anything that would bring me to another part of the site would not work (even clicking the "previous page" button). I was still able to switch tabs and navigate other websites.
I don't think this is a really concerning issue, but it could lead some people to believe the site has crashed.
I was using google chrome on windows 10, on the leed's university's eduoram wifi.

Copied from original issue: #198

When logging into this repo, I get the security alert regarding the "randomatic" dependency. It should have at least version 3.0.0 to solve this issue.

From @jakobbossek on September 19, 2016 0:42

What I (generally) really dislike - not only on the OpenML website - is opening external links in the same tab/window. Personally I hate that. I highly prefer them to be opened in a new tab/window via taregt=_blank or via javascript, e.g., ...onclick="javascript:window.open('myURL', ...);". Moreover, I like external links to have atitlestarting withExternal link - Website of ...`.

What is your opinion?

Copied from original issue: #65

From @amueller on October 10, 2017 21:23

For classification datasets with, say <10 classes, I think it would be great to show not a single box-plot for continuous features, but a boxplot by class. That tells you immediately which features are valuable, and you can also see if there is data leakage.

Happy to give this a shot if someone (@janvanrijn ?) tells me where to look for this.

Copied from original issue: #193

From @joaquinvanschoren on April 4, 2017 17:53

E.g. get all runs on task X, ordered by AUC

Copied from original issue: #121

From @HeidiSeibold on October 12, 2017 12:26

Dataset(s)

What do I have to put in for dataset(s)? The id, the name, the url?

Target feature

Would it be possible to have a drop down menu from all possible variable names?

Estimation procedure

Should we have 10 times 10 CV as default? We should give a reason why choos the default. Maybe there should be a link with info on when which type of CV is useful.

Evaluation measures

Could we have a drop down menu for all evaluation measures that are useful for this task type?

Copied from original issue: #196

From @amueller on November 23, 2016 16:31

I tried to reset my password.
Using a simple password I got an error "Please match the requested format." but the format is not described on that page, so I have no way of knowing what I did wrong.

Copied from original issue: #75

From @amueller on November 16, 2017 17:38

It looks like the page has no maximum width setting, which can lead to very long lines in the wiki for datasets or the user guide.
Generally long lines make reading text harder.Check out how this tensorflow tutorial is fixed width: https://www.tensorflow.org/tutorials/wide_and_deep (though actually quite wide).

Compare the to the NYT https://www.nytimes.com/2017/11/16/us/politics/house-tax-overhaul-bill.html?hp&action=click&pgtype=Homepage&clickSource=story-heading&module=first-column-region&region=top-news&WT.nav=top-news
which basically keeps most of the page blank (or, like the github issue tracker).

Not entirely sure where this container is defined, so opening an issue instead of trying to track it down.

Copied from original issue: #211

From @amueller on November 9, 2017 19:49

I searched for the "collins" dataset:
https://www.openml.org/search?q=collins&type=

The first result is an inactive version of the dataset with no runs, and it's followed by tasks with no runs.
If I select "datasets" on the right, I get a different dataset on top, I think.

Copied from original issue: #205

From @joaquinvanschoren on September 20, 2017 7:21

The dataset upload process via the website could be smoother:

• Step 1: Upload a dataset (only file/link and name). Also point to a tutorial to upload data via the APIs
• Step 2: Data is analyzed, show the web page with the (wiki) description and features. Qualities are computed in the background and show up a bit later. On that page, people can choose license, add full description, set special features, upload a new version.
• Step 3: Data checking bot checks whether the dataset can be made active, if not, show reasons why not.

Arguably, dataset upload via the Python/R/Java API can happen in the same way, step 1 is then done in code.

Copied from original issue: #183

From @HeidiSeibold on October 11, 2017 9:32

I first went to the measures and filtered for evaluation measures, i.e.:

https://www.openml.org/search?q=%2520measure_type%3Aevaluation_measure&type=measure

when I then clicked on task type, I go to:

https://www.openml.org/search?type=task_type&q=%20measure_type:evaluation_measure

which obviously does not show me all task types.

Copied from original issue: #194

From @mfeurer on May 24, 2017 8:33

As the scikit-learn API creates horribly long flow names, it would be great if the interface could display the full name in an overlay textbox when hovering over the flow (like it displays the trash bin):

Copied from original issue: #140

From @HeidiSeibold on January 10, 2017 14:30

I found two issues with a data set and wanted to submit them.

http://www.openml.org/d/1513

But somehow I can not submit a second issue. The screen looks like this:

For other data sets it looks like this

Copied from original issue: #83

When creating a task through the webinterface (by clicking define a new task button at the bottom at a dataset page), you have to wait for ~5 seconds after the page is shown before selecting the task type in the dropdown menu. Otherwise, the item selected in the dropdown menu will be ignored.

From @HeidiSeibold on March 13, 2017 11:34

From the task websites
https://www.openml.org/t/4850
and
https://www.openml.org/t/2291
I can not see the difference.

By following the link to the data I can figure out that the difference is the data set version.

What I think would be nice, is if there were a short description of the data set, estimation procedure, evaluation measures, target feature, etc

Copied from original issue: #108

From @joaquinvanschoren on March 2, 2017 17:12

What I currently can't do:

• for specific study, plot all performances of all algorithms on a dataset (or one algorithm on all datasets)

This is possibly solved by having a study and tag filter in the dataset/flow search

Copied from original issue: #101

From @thvasilo on January 30, 2018 12:16

In the dataset search field, enter a few characters to get the auto-complete drop-down list.

Navigate to the desired dataset using the arrow keys on the keyboard.

Expected: Press enter to navigate to the highlighted dataset.

Result: Enter simply closes the dropdown list, it's necessary to click with the mouse to navigate to dataset from the dropdown.

Copied from original issue: #224

I tried running the website locally and got

0 info it worked if it ends with ok
1 warn npm npm does not support Node.js v7.6.0
2 warn npm You should probably upgrade to a newer version of node as we
3 warn npm can't make any promises that npm will work with this version.
4 warn npm Supported releases of Node.js are the latest release of 6, 8, 9, 10, 11.
5 warn npm You can find the latest version at https://nodejs.org/
6 verbose cli [ '/usr/local/bin/node', '/usr/local/bin/npm', 'run', 'debug' ]
7 info using [email protected]
8 info using [email protected]
9 verbose run-script [ 'predebug', 'debug', 'postdebug' ]
10 info lifecycle [email protected]~predebug: [email protected]
11 info lifecycle [email protected]~debug: [email protected]
12 verbose lifecycle [email protected]~debug: unsafe-perm in lifecycle true
13 verbose lifecycle [email protected]~debug: PATH: /usr/local/lib/node_modules/npm/node_modules/npm-lifecycle/node-gyp-bin:/home/andy/checkout/website-1/node_modules/.bin:/home/andy/anaconda3/envs/py37/bin:/home/andy/gems/bin:/home/andy/.npm-global/bin:/home/andy/.gem/ruby/2.3.0/bin:/home/andy/.local/bin:/home/andy/anaconda3/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin
14 verbose lifecycle [email protected]~debug: CWD: /home/andy/checkout/website-1
15 silly lifecycle [email protected]~debug: Args: [ '-c', 'webpack -d' ]
16 silly lifecycle [email protected]~debug: Returned: code: 1  signal: null
17 info lifecycle [email protected]~debug: Failed to exec debug script
18 verbose stack Error: [email protected] debug: `webpack -d`
18 verbose stack Exit status 1
18 verbose stack     at EventEmitter.<anonymous> (/usr/local/lib/node_modules/npm/node_modules/npm-lifecycle/index.js:301:16)
18 verbose stack     at emitTwo (events.js:106:13)
18 verbose stack     at EventEmitter.emit (events.js:192:7)
18 verbose stack     at ChildProcess.<anonymous> (/usr/local/lib/node_modules/npm/node_modules/npm-lifecycle/lib/spawn.js:55:14)
18 verbose stack     at emitTwo (events.js:106:13)
18 verbose stack     at ChildProcess.emit (events.js:192:7)
18 verbose stack     at maybeClose (internal/child_process.js:890:16)
18 verbose stack     at Process.ChildProcess._handle.onexit (internal/child_process.js:226:5)
19 verbose pkgid [email protected]
20 verbose cwd /home/andy/checkout/website-1
21 verbose Linux 4.15.0-36-generic
22 verbose argv "/usr/local/bin/node" "/usr/local/bin/npm" "run" "debug"
23 verbose node v7.6.0
24 verbose npm  v6.4.1
25 error code ELIFECYCLE
26 error errno 1
27 error [email protected] debug: `webpack -d`
27 error Exit status 1
28 error Failed at the [email protected] debug script.
28 error This is probably not a problem with npm. There is likely additional logging output above.
29 verbose exit [ 1, tru

any idea what the issue is?

From @amueller on March 14, 2016 11:45

I think there should be a nicer error message (though this is clearly not high priority)

Copied from original issue: #40

From @jakobbossek on September 19, 2016 0:36

Some of the "iconic" links do not have a title attribute assigned. Links should always have an assigned title! Moreover, on classical "mouse-based" operating systems, CSS tooltips are a valuable advance over archaic HTML tooltips. The latter look bad and appear with a > 1s delay.
Since you are using jquery, you might have a look at my minimalistic tooltip lib

Copied from original issue: #64

From @HeidiSeibold on December 9, 2016 15:26

My download activity seems too low. Only in the last few days I have downloaded many tasks (way more than 15).

Copied from original issue: #80

From @mfeurer on May 24, 2017 8:33

As the scikit-learn API creates horribly long flow names, it would be great if the interface could display the full name in an overlay textbox when hovering over the flow (like it displays the trash bin):

Copied from original issue: #140

From @Moanraker on November 23, 2016 10:41

An account without a registered First Name will cause the headers of some pages to display weird titles.

Copied from original issue: #74

From @thvasilo on November 28, 2017 9:42

This is some behavior I've noticed on FF57:

1. I click "Like" on a dataset.
2. Then I click the "Download" button

This results in the like being removed, is this intentional?

Copied from original issue: #217

From @DraXus on February 15, 2015 17:20

I think it would be useful to include additional filters in the tasks search page such as:

• Number of instances
• Number of features
• Number of missing values
• Number of classes

So those filters allow to look for tasks that use datasets of such criteria.

Copied from original issue: #18

The markdown conversion on study and data description pages seems to be broken. The 'Loading wiki' message does not disappear and only raw markdown is shown. This makes it impossible to edit the text, as the edit button appears only after the wiki is loaded. I've had this problem since last Thursday I think (but then I assumed it had something to do with the heavy server load).

From @mfeurer on November 10, 2017 16:23

This image shows 101 datasets for study 14:

If I click on 'SEARCH THESE DATA SETS IN MORE DETAIL':

it thinks that there are only 99 data sets.

Copied from original issue: #206

From @HeidiSeibold on October 4, 2017 11:16

I just created a task. This is what I see:

Would be nice to see a confirmation that the task was successfully created and what the task id is (with a link).

Copied from original issue: #189