Sometimes you see something on W E ∆ T H E R that you want to check out and maybe buy, click on the image to buy it … and it’s not there.
It’s a total bummer – we feel you.
The issue is that the products and their associated shopping links come from Svpply, and as stores update their websites these links stop working. As of now, there’s (sadly) not a whooole lot we can do about it.
Would it be useful to have the name of the product (when available)? At least you could do some internet investigating on your own if you really wanted it.
Nate Silver, whom I love (especially now that it’s election season), on weather forecasting and probability:
“It was enlightening to speak with men and women at the forefront of science and technology. But I found that despite their best efforts, their predictions have often gone poorly.
“The discipline of meteorology is an exception. Weather forecasts are much better than they were 10 or 20 years ago.”
Check it – this graph is traffic to W E ∆ T H E R over the past month:
Sometime on August 27 is when our old weather API got pulled. You can see right there (there!) on September 14 when Jack saved everything and the flat line starts going back up.
What got me curious is that almost immediately after the site went back up I started seeing tweets about it again; it was like people were checking the site regularly. I looked back at a day with normal traffic and sure enough, a lot of users turn out to be return visitors.
New vs. returning visitors – Sept. 20:
Why’d I go back to a day with normal traffic? Well today when I went to get some numbers for this post I noticed that the percentage seemed closer to 50/50 than usual.
New vs. returning visitors on Sept. 26:
This also seems to be proven anecdotally:
I thought this little project translating data into human experience would just be a fun experiment, so I’m absolutely delighted to see that it’s actually useful to people. Awesome.
After a small hiccup in service due to the loss of our weather API, we’re back (mostly). We missed you!
A small note for international users: you might have noticed that W E ∆ T H E R no longer works outside the US – our apologies. We’re looking for a (free) weather API that works worldwide, so if you know of one do let us know.
Thanks for understanding as we try to stay out of the rain here (metaphorically).
Yesterday the API we use for getting weather data stopped working, sadly. It appears that it’s been shut down for good, but even if it’s just down for now we’re still shopping for a new API.
It’s hard to say when we’ll be back up, but rest assured we’re working on it (and by we, I mostly, and guiltily, mean Jack). I’m also hoping we can use this down-time and the necessary rebuilding that will be required to come back with some new features and fixes.
This is also a good reminder: if your product relies on external services – and most do – have a contingency plan for when (not if) you need to find an alternative. APIs shut down, startups fail, terms of service change; don’t let your product be at the mercy of these external conditions.
Thanks for your patience. See you all soon.
I previously wrote a note about recommendations and addressing the fact that they aren’t always accurate. I wanted to follow up and briefly explain why.
The main problem comes from weather conditions that are incongruous with common expectations of accompanying temperatures. For example, it’s very easy to recommend shorts when it’s hot (temperature) and sunny (condition) out. We tend to think of this as fairly common summer weather. It’s more difficult, however, to recommend something when it’s hot and also hailing, as it was last week in New York. A raincoat would be a reasonable recommendation from a computer, but a human would know that’s going to be a bit uncomfortable. We can’t automate “wait inside until the summer squall passes.”
Addendum: A related difficulty with recommendations is that during many days there can be a sizable swing between the high and low temperatures, (e.g. hot days and cool nights). You can offer some recommendations for the high and some for the low, or you can take the average of the two and base your recommendations on that. Alternatively you could get the current temp and base the recommendations on that, which would be somewhere between the high and low. None of these methods is perfect, and some are more difficult to implement than others.
A secondary problem is simply that clothes don’t come with guidelines for when it’s appropriate to wear them. The products are picked from Svpply and put into buckets based on temperature and condition. Sometimes my picks are off, and sometimes these buckets are just too broad to be perfectly relevant to the conditions outside your window.
The solution is to find a balance between weighing temperature and condition, and we’re obviously not there yet. But we’re working on it.
When we first launched, Jack and I talked about whether your location should be automatically detected when you go to wevther.com or launch the web app. While it’s ideal to immediately see your own city, we decided that it was a bit intrusive to be asked for permission to find your location right away. Ultimately we decided that we’d show a few default cities and you could change your location manually.
Starting today, though, we’re going to test auto-detecting your location when you launch the site. After hearing from users, it seems like automatically showing the correct location might trump the slight annoyance of being confronted with the dialog box asking if you want to let it locate you.
Let us know what you think here or @twitter.