@Halfak which classifier algorithm should I use? Current production algorithms or HashingVector + GradientBoosting?

Also, regarding ROC score difference between T128087#2600696 and current: current one is correct. GridSearchCV is calculating it. Earlier I was first calculating myself using roc_auc_score.

Plots with and without sample weights:

Agreed. I'm on it.

Here is the plot.

Makes sense. Will work on it.

Here are the results from the grid search:

Best ROC AUC Score:
0.910445174634

Here is the link to compare the results against:

Will try it

New results with below params:

@Halfak: results are out. Here you go:

I am also pretty close to evaluating the first model with combined
features😃.

I think I got it wrong about the threshold in my above comment. Now I get it as the threshold is to keep the features whose importance is >= threshold, right? 0.0005 yields 375 features. I updated the notebook to show these features.

Shouldn't it be (2^20)*0.0005? It is 524.

@Halfak I've plotted the histogram of the GBC (without other 77 features). Here is the exported version of the ipython output. I am exporting it and attaching here to capture the historical snapshot. Do you think there is an obvious step at 0.0005 (three zeros before five that is).

Closed by mistake.

For the record, the IRC log is here:

Here is a pull request using the first approach:

The following answer solves the problem:

I've found the reason behind the issues. The expression var browserLanguages = navigator.languages || [ navigator.language || navigator.userLanguage ]; assigns 'en-us' in iOS/Mac safari, but in Chrome, it assigns 'en'. en-us could not be found in the sourceLanguages hence it becomes undefined by the following filter:

@schana replied back

@schana: pls look into the PR and share your thoughts there

Here is a first working copy:

Questions: what will be the behavior when I click on a search suggestion? Should it do anything other than filling up the search text input with the title of the suggestion?

Yes. It still exists. Checked both in iOS Chrome and Safari.

@Jdrewniak there are dependencies to addEvent, mw, getDevicePixelRatio. Could you pls point me to the right sources for getting these?

@Jdrewniak Thanks. Let's go with the first one. An example of it being used would be very helpful.

Is the Wikipedia search suggestion box available as widget/extension/lib to integrate? Do we have to build one of our own?

@schana - I've updated the pull request to fetch bigger images (512px). Please LMK if it is good enough for the merge.

I can see that some articles do have images but the query API does not return any. What could be the reason? Example:

In that case, I think 250px for width is good enough; as the size of a card size is around 248px X 128px.

I did a quick check. If I open https://recommend.wmflabs.org/, all requests (including XHR, static) are made to https://. So, are we safe?

Pginer-WMF - Also, would like to hear your opinion on images getting
pixelated.

I like it. What about others? Do need approval from design team?

I think it will help if we can have a predictable way to way to recreate the situation. Where exactly can we hardcode and always return articles with '/'?

I can confirm in OS X, Chrome it is red.

Here is a pull request:

Can we have an image like this: http://www.clker.com/clipart-15263.html

Here is the pull request:

I tried this.

@JustinOrmont, @Halfak: I have few questions:

I have created a pull request:

I could think of two ways to achieve this.

@JustinOrmont Thank you for the detailed input, appreciate :)

Here are the new scores with n_estimators=200 and learning_rate = 0.05

How are you parameterizing the GradientBoostingClassifier model when constructing it? I think that it would be worthwhile to try large values for n_estimators and small values for learning_rate. I'm not sure what kind of effect this will have so it will certainly be an experiment.

There were few discussions over email. I'm pasting that here for future reference. It is in reverse chronological order.

In T106638#2091356, @Halfak wrote:

@schana interested in this task? It seems like setting this up in puppet makes sense. Here's the command we run to start the sync server:
ores precached stream.wikimedia.org https://ores.wmflabs.org --config ores-wikimedia-config/config/00-main.yaml

@Halfak Please review when you get a chance. Should I be assigning this card to you (the reviewer)? Not sure what workflow we follow here.

Here is the commit for the change:

I have coded the part to install systemd service. I have craeted a new class using wikitech "manage puppets group". It is called role::labs::ores::precached.

Note to self from a chat with yuvi:

Is anyone working on this? If not, I'll take it up.