Add keras example API for an autoencoder model #834
Conversation
|
This looks really awesome, thank you for contributing this! I did a quick pass, and everything looks good on first glance. I will do a more thorough review in the next day or two, and I'll upload the the model to the cortex examples bucket and try running the API. I'll keep you posted on how it goes. |
|
|
||
| # encode image | ||
| image_enc = base64.b64encode(byte_im).decode("utf-8") | ||
| image_dump = json.dumps({"img_pred": image_enc}) |
There was a problem hiding this comment.
I don't think you need to do a json.dumps here. You should be able return {"img_pred": image_enc}. Then in https://github.com/cortexlabs/cortex/pull/834/files#diff-4e5041cf7620e33613b081c648188d95R66, you wouldn't todo a json.loads
There was a problem hiding this comment.
Hmm, I can try this.
|
|
||
| ## Sample Prediction | ||
|
|
||
| ### Dependencies |
There was a problem hiding this comment.
The dependencies section can be removed since only curl, sed and base64 are being used right?
There was a problem hiding this comment.
This one slipped through. Thanks! It's going out.
| ```bash | ||
| curl "${ENDPOINT}" -X POST -H "Content-Type: application/json" -d '{"url":"'${IMAGE_URL}'"}' | | ||
| sed 's/"//g' | | ||
| base64 -do prediction.png |
There was a problem hiding this comment.
Although the -o flag is an option in OSX, I don't think it is an option in ubuntu's base64 binary. To be on the safe side I recommend you replace base64 -do predictor.png with base64 -d >> predictor.png
There was a problem hiding this comment.
Hmm, I agree. I'll do the appropriate modification.
This PR contains an example project built with Keras.
The model that I have implemented is an autoencoder. In this use case, the autoencoder is used to filter out the noise from text documents. The
cortex.yamlconfig file uses the CPU for inference.Here's a before-and-after situation:
The model has to be uploaded to
s3://cortex-examples/keras/document-denoiserdirectory. Its name has to bemodel.h5. The model can be found here (temporarily).For the loading function of the model I had to resort to importing it this way
Instead of
Because I encountered the error specified in #831. This workaround is good for both types of hardware deployments: CPU or GPU.
checklist:
make testandmake lintsummary.md(view in gitbook after merging)