beat issueshttps://gitlab.idiap.ch/groups/beat/-/issues2014-11-21T12:02:48Zhttps://gitlab.idiap.ch/beat/beat.web/-/issues/74Error reading parameter for analysis block2014-11-21T12:02:48ZManuel GüntherError reading parameter for analysis blockI continue receiving "internal BEAT system error"s while trying to access a parameter in an analysis block. The relevant code is something like:
```python
## INPUTS
## scores: tutorial/probe_scores/1
## INPU...I continue receiving "internal BEAT system error"s while trying to access a parameter in an analysis block. The relevant code is something like:
```python
## INPUTS
## scores: tutorial/probe_scores/1
## INPUTS_END
## RESULTS
## ROC: beat/chart/1
## RESULTS_END
## PARAMETERS
## points: uint16 100
## PARAMETERS_END
import bob
class Algorithm:
def __init__(self):
self.positives = []
self.negatives = []
def setup(self, parameters):
self.points = parameters.get('points', 100)
return True
def process(self, inputs, output):
...
roc = bob.measure.roc(self.negatives, self.positives, self.points)
...
return True
```
(see https://www.beat-eu.org/platform/algorithms/siebenkopf/ROC/6, if you can). Second BEAT Reviewhttps://gitlab.idiap.ch/beat/beat.web/-/issues/75Cannot use same analysis tool twice2014-11-21T10:53:29ZManuel GüntherCannot use same analysis tool twiceThe short verison: When I have two analyzers that output the same name, only one of it is shown in the result section.
The longer version:
I have a complicated toolchain, including dev and eval sets.
In total, I have three analyzers...The short verison: When I have two analyzers that output the same name, only one of it is shown in the result section.
The longer version:
I have a complicated toolchain, including dev and eval sets.
In total, I have three analyzers, one computing EER and HTER from both sets, and an ROC curve generator for each of the sets.
In my experiment, I use the simple algorithm to plot the ROC curves for both the dev and eval set. I expected two plots. However, since the output of the ROC algorithm is called "ROC" and there is no way to change this name in the toolchain, I get only one ROC curve displayed (I guess the one that was produced first).
See: https://www.beat-eu.org/platform/experiments/siebenkopf/FaceRec-NoTraining/1/PhaseDiffPlusCanberra_TT_Banca_P8/ if you can.
https://gitlab.idiap.ch/beat/beat.web/-/issues/76Cannot reset database configurator2015-06-17T15:02:41ZLaurent EL SHAFEYCannot reset database configuratorIt is currently not possible to reset the database configurator from its discrete state into the database protocol selector format. It would be nice to have a reset button to allow the user to do so on the experiment confiugration page.It is currently not possible to reset the database configurator from its discrete state into the database protocol selector format. It would be nice to have a reset button to allow the user to do so on the experiment confiugration page.BEAT Tourhttps://gitlab.idiap.ch/beat/beat.web/-/issues/77Backend page time out since partitions are now -noac mounted2014-12-02T12:12:52ZLaurent EL SHAFEYBackend page time out since partitions are now -noac mountedThe backend page now seems to time out when e.g. a clean-up is performed.
This is caused by the use of the -noac NFS flag while mounting the partition on beatsched, which affects the filesystem performance.The backend page now seems to time out when e.g. a clean-up is performed.
This is caused by the use of the -noac NFS flag while mounting the partition on beatsched, which affects the filesystem performance.Second BEAT Reviewhttps://gitlab.idiap.ch/beat/beat.web/-/issues/78Locked attestations do not properly expire2015-11-20T11:36:56ZAndré AnjosLocked attestations do not properly expireThe `creation_date` is correctly set for all locked attestations. The `publication_date` is set for all public attestations. The only thing we need to implement to get this going is:
- [x] A warning on the locked attestation page that t...The `creation_date` is correctly set for all locked attestations. The `publication_date` is set for all public attestations. The only thing we need to implement to get this going is:
- [x] A warning on the locked attestation page that tells the user about the remaining unlock time (which is a constant on the `settings.py` file for the project.
- [x] A cronjob that runs a (very simple) script that, for all attestations w/o publication_date, compares the current date to the creation date and deletes that attestation (just the attestation, not the experiment) if the time difference pass a threshold.Final Documentationhttps://gitlab.idiap.ch/beat/beat.web/-/issues/79Cancelled experiment page is wrong when no associated jobs was cancelled2014-11-20T14:49:24ZLaurent EL SHAFEYCancelled experiment page is wrong when no associated jobs was cancelledThe cancelled experiment page does not show any useful information, when the cancellation of this experiment does not involve the cancellation of a specific job. In this case, we get the configuration page for this experiment instead of ...The cancelled experiment page does not show any useful information, when the cancellation of this experiment does not involve the cancellation of a specific job. In this case, we get the configuration page for this experiment instead of a page indicated that the experiment has been cancelled.
I do not know the exact cause of the problem. I suspect that the lack of error message such as 'Job x was cancelled by an external user agent' leads to the problem.Second BEAT ReviewLaurent EL SHAFEYLaurent EL SHAFEYhttps://gitlab.idiap.ch/beat/beat.web/-/issues/80validate_queue_environment() does not work well2014-11-21T11:50:14ZLaurent EL SHAFEYvalidate_queue_environment() does not work wellWhile performing some update on the database, we noticed that validate_queue_environment() was not validating a tuple (Queue, Env), which seemed to be correct. We have to investigate why.
https://gitlab.idiap.ch/biometric/beat.web/blo...While performing some update on the database, we noticed that validate_queue_environment() was not validating a tuple (Queue, Env), which seemed to be correct. We have to investigate why.
https://gitlab.idiap.ch/biometric/beat.web/blob/master/beat/web/experiments/models.py#L35Second BEAT Reviewhttps://gitlab.idiap.ch/beat/beat.web/-/issues/81Tabbed scheduler page2014-11-24T12:54:34ZAndré AnjosTabbed scheduler pageCurrently, the scheduler information is just output in a single page. Following other functionality available in the web site, it would be nice to organize this information in a tabbed view.
The default tab to show would be a table, e...Currently, the scheduler information is just output in a single page. Following other functionality available in the web site, it would be nice to organize this information in a tabbed view.
The default tab to show would be a table, emitted by the scheduler with the current experiments running and their stats (top-like). The other tabs would concentrate information about the "workers" and "queues".Second BEAT Reviewhttps://gitlab.idiap.ch/beat/beat.web/-/issues/82Harmonisation of statistics and processing times sent by the scheduler2015-11-20T11:35:02ZAndré AnjosHarmonisation of statistics and processing times sent by the schedulerWe must harmonize the displaying of computing power informed by the scheduler and calculated by the worker in some way. Currently, we find some inconsistencies between the values displayed on the user page (experiments tab) and the "sequ...We must harmonize the displaying of computing power informed by the scheduler and calculated by the worker in some way. Currently, we find some inconsistencies between the values displayed on the user page (experiments tab) and the "sequential processing" times displayed on the experiment page.Final Documentationhttps://gitlab.idiap.ch/beat/beat.web/-/issues/83Django FloatField does not handle +/-inf and NaN values2015-04-27T09:58:37ZLaurent EL SHAFEYDjango FloatField does not handle +/-inf and NaN valuesThe current version of Django (1.7.1) has a FloatField implementation that does not support special floating point values such as +/-inf and NaN.
The following Django ticket mentions this issue:
https://code.djangoproject.com/ticket/...The current version of Django (1.7.1) has a FloatField implementation that does not support special floating point values such as +/-inf and NaN.
The following Django ticket mentions this issue:
https://code.djangoproject.com/ticket/4287
This is currently a problem with our analyzers that store results as FloatField in the SQL database.
When such a value is passed from the scheduler to the web (e.g. in the block_finished method), an error occurs (500: internal server error), which leads to the messsage to be continuously sent and misprocessed.
We need to find a workaround for this.Tutorial for the FG2015 ConferenceSamuel GAISTSamuel GAISThttps://gitlab.idiap.ch/beat/beat.web/-/issues/84Synchronisation issue between the system status and the internal status ?2015-03-27T06:09:52ZSébastien MARCELSynchronisation issue between the system status and the internal status ?there might be a synchronisation issue between the system status ( https://www.beat-eu.org/platform/ ) and the internal status. Indeed, the status says that 1 experiment is running but clearly looking at the scheduler page this is not th...there might be a synchronisation issue between the system status ( https://www.beat-eu.org/platform/ ) and the internal status. Indeed, the status says that 1 experiment is running but clearly looking at the scheduler page this is not the case.
Tutorial for the FG2015 ConferenceAndré AnjosAndré Anjoshttps://gitlab.idiap.ch/beat/beat.web/-/issues/85CPU time on expe page is wrong ?2015-02-05T09:57:30ZSébastien MARCELCPU time on expe page is wrong ?
I ran a new quick expe
https://www.beat-eu.org/platform/experiments/smarcel/elie_khoury/ubmgmm_voxforge/1/test7/
which is a fork of
https://www.beat-eu.org/platform/experiments/smarcel/elie_khoury/ubmgmm_voxforge/1/ubmgmm-voxforge...
I ran a new quick expe
https://www.beat-eu.org/platform/experiments/smarcel/elie_khoury/ubmgmm_voxforge/1/test7/
which is a fork of
https://www.beat-eu.org/platform/experiments/smarcel/elie_khoury/ubmgmm_voxforge/1/ubmgmm-voxforge-g1024iter100-rr/
but with much less gaussians such as the execution is fast.
After expe "test7" is completed I looked at the main expe page: https://www.beat-eu.org/platform/user/smarcel/?tab=experiments
where I can see that CPU time is 23.2 minutes. This is obviously wrong this is was VERY fast as it can be observed by inspecting the sequential execution time for each block on
https://www.beat-eu.org/platform/experiments/smarcel/elie_khoury/ubmgmm_voxforge/1/test7/
https://gitlab.idiap.ch/beat/beat.web/-/issues/86Missing API for downloading databases and views2015-03-18T12:16:12ZAndré AnjosMissing API for downloading databases and viewsWe're currently missing an API to download databases (JSON and views) associated with experiments. There should be no reason why this should not be possible.
Currently, after downloading and experiment, the toolchain and related algor...We're currently missing an API to download databases (JSON and views) associated with experiments. There should be no reason why this should not be possible.
Currently, after downloading and experiment, the toolchain and related algorithms and dataformats, the user may not validate the experiment because it is lacking the database definition.
Clearly, we're not talking about the real data, just the (JSON and python views) associated with the platform. Confidentiality should be respected in all cases.Tutorial for the FG2015 ConferenceAndré AnjosAndré Anjoshttps://gitlab.idiap.ch/beat/beat.web/-/issues/88Implement support for beat.core description/documentation API2015-05-06T07:45:26ZAndré AnjosImplement support for beat.core description/documentation APIThe new beat.core contains direct support for object descriptions and a documentation API.
Following the other functionality that is already well-merged with beat.core, it would be good that this package also used the beat.core APIs t...The new beat.core contains direct support for object descriptions and a documentation API.
Following the other functionality that is already well-merged with beat.core, it would be good that this package also used the beat.core APIs to set and get object descriptions and documentation.
The object `short_description` in beat.web (known as `description` in beat.core), can still be cached inside beat.web, but should be updated on beat.core once it changes. This allows users to simply edit the object JSON files and still be able to push the description to the server, when they are using the command-line interface.
Similarly, `description` in beat.web (known as `documentation` in beat.core) should be materialized in a proper file, like it is done for the JSON declarations. This would allow users from the command-line to edit this file and push it to the server.Tutorial for the FG2015 ConferenceAndré AnjosAndré Anjoshttps://gitlab.idiap.ch/beat/beat.web/-/issues/89Django experiments/Results table could be largely simplified2015-05-04T13:13:37ZAndré AnjosDjango experiments/Results table could be largely simplifiedCurrently, our Results table contains 4 columns that store data types: one for integers, one for floats, one for booleans and one for "everything else". In order to know how to read the data, the platform consults the value of the "type"...Currently, our Results table contains 4 columns that store data types: one for integers, one for floats, one for booleans and one for "everything else". In order to know how to read the data, the platform consults the value of the "type" column" and then reads the corresponding field, with an adequate de-serializer.
In order to support the new plotting standard, the choice of implementation would be to add a "__type__" field on chart data (saved at the "Data value" column), which implies in a hack to ignore such a field on beat.core.
The way to solve this cleanly could go through:
1. Re-write the Result table so that all results are represented by strings or binary blobs
2. Condition the readout of Result tables to the type of the data object, as defined in the "type" column, which would have to be set to a "string" type.
This technique would remove 3 columns from the table, simplifying it, at the cost of reviewing the code that uses/displays results.
Because all results would be expressed into string (or binary), no need to worry about issue #83 anymore, would this solution be taken.Tutorial for the FG2015 ConferenceLaurent EL SHAFEYLaurent EL SHAFEYhttps://gitlab.idiap.ch/beat/beat.web/-/issues/90Harmonisation of experiments web API2015-05-13T15:25:18ZAndré AnjosHarmonisation of experiments web APIWe presently have our web API for experiments in a bit of an inconsistent state:
1. Some of the calls go through <base>/experiments
2. Some others go through <base>/experiments/configuration
This is not the case for other platform...We presently have our web API for experiments in a bit of an inconsistent state:
1. Some of the calls go through <base>/experiments
2. Some others go through <base>/experiments/configuration
This is not the case for other platform objects (such as data formats, toolchains, or else). This creates a slight complication for API users.
@philip.abbet: is this just legacy behaviour or is there a good reason for keeping it this way?API DocumentationSamuel GAISTSamuel GAISThttps://gitlab.idiap.ch/beat/beat.web/-/issues/91DELETE action was removed from <base>/experiments/configuration/...2015-03-27T12:38:52ZAndré AnjosDELETE action was removed from <base>/experiments/configuration/...According to @samuel.gaist, this was done during the rest framework port. We should bring this API back.According to @samuel.gaist, this was done during the rest framework port. We should bring this API back.Tutorial for the FG2015 ConferenceSamuel GAISTSamuel GAISThttps://gitlab.idiap.ch/beat/beat.web/-/issues/92Team creation form2015-03-27T12:37:48ZAndré AnjosTeam creation formThe team creation form is in place. A couple of fixes are due:
- [x] It would be good that we describe, somewhere, what the different privacy levels mean.
- [x] It would be better to have the whole form completable at once, in a sing...The team creation form is in place. A couple of fixes are due:
- [x] It would be good that we describe, somewhere, what the different privacy levels mean.
- [x] It would be better to have the whole form completable at once, in a single page, instead of creating the team and then editing its attributes
- [x] It would be good if, by clicking on the team (at the user micro-site), the system took the user to the team edition page as for other objects. Some harmonization should be put in place for this.Tutorial for the FG2015 ConferenceSamuel GAISTSamuel GAISThttps://gitlab.idiap.ch/beat/beat.web/-/issues/94URL field "format=string" does not work after Django rest framework port2015-03-27T12:38:34ZAndré AnjosURL field "format=string" does not work after Django rest framework portThis field in the URL is used by the rest framework to define which serializer/marshaller to use.
We will need to change this everywhere to something like `object-format=string`.This field in the URL is used by the rest framework to define which serializer/marshaller to use.
We will need to change this everywhere to something like `object-format=string`.Tutorial for the FG2015 ConferenceSamuel GAISTSamuel GAISThttps://gitlab.idiap.ch/beat/beat.web/-/issues/95Search for experiments with datasets of different databases2015-06-17T07:04:23ZTiago de Freitas PereiraSearch for experiments with datasets of different databasesHi,
In the experiment setup, it is possible to play with datasets of different databases.
For example, I can define the train and the development set of one database and select the test set of another.
However, once the experiment i...Hi,
In the experiment setup, it is possible to play with datasets of different databases.
For example, I can define the train and the development set of one database and select the test set of another.
However, once the experiment is done, there is not a nice way to serch for this particular dataset setup.
I think we should think in a way to do it nicely.
Thanks
TiagoBEAT Tour