cc-ai / floods-backend Goto Github PK

CirceCI is failing for a few reasons, let's keep this passing so we don't miss any real errors that make it through.

@jsnrdl can you change all references to CCAI to VICC? (and Climate Change AI to Visualizing the Impacts of Climate Change)

We'll change the org too

Right now, an address that has no climate metadata available returns something like:

  "metadata": {
    "monthly_average_precipitation": {
      "title": "Monthly Average Precipitation in 2050", 
      "value": NaN
    }, 
    "relative_change_precipitation": {
      "title": "Relative Change in Precipitation by 2050", 
      "value": 0.0
    }
  }, 

But this is not valid JSON. In cases like this, let's return "value": null instead.

For now, we are only going to apply the generative model to the image if the relative change in precipitation that is expected in 2050 is above 60%.

Right now, when the webserver gets deployed, existing pods get terminated before new pods are ready. We need to configure rolling deploy configurations and readiness probes to ensure deploys are zero-downtime.

if the address doesn't have a street view, for example "38 Galshoyan Street, Yerevan, Armenia"
or it is a wrong address, we are getting an error
Image processing error: Error: Request failed with status code 500

also, if previous request was successful, it keeps the title and street view of the previous one. Should be empty in case of the error, to do not confuse the user

Instead of the api.py file.

We need to indicate that images returned from the API are for testing and development only. For this reason, apply a watermark to every image that uniformly makes it clear that this is the case.

In production, it will be useful to understand the regions where people are interested in climate change as well as the efficacy of our models. For this reason, we need to log the requested address, the raw image we get from streetview, and the flooded image.

Let's also make sure that we have our legal and privacy ducks in a row before implementing this. A set of terms from a website like Montreal Data License might be useful for this.

Figure out why torch.cuda.is_available() returns False when running via a CUDA base container on a Kubernetes node that has a NVIDIA P100 available.

The files are too big for Git LFS now so let's pull them down potentially via an init container.

For some places, everything works just fine:

Now, if I query 6666 rue saint Urbain in Montreal, the following error is returned:

However, if I query 6666 rue saint Urbain and it is the FIRST address I am querying when logging into th website, the error is not returned, but no image is shown: