Gramex 1.67 release notes

Gramex 1.67 introduces MLHandler (to train, deploy and re-train models) and simpler names for micro-services.

Train ML models

MLHandler helps you embed ML models that can be trained on user-provided data.

You can embed a scikit-learn, statsmodels or tensorflow/keras model. For example:

url:
  classifier:
    pattern: /rugby
    handler: MLHandler
    kwargs:
      model:
        class: SVC # Embed sklearn.svm.SVC
        target_col: ans # Predict the "ans" column

You can send HTTP requests to /rugby to teach it that tall heavy boys make the male rugby team. Suppose this data is in data.csv:

| sex | height | weight | ans |
|-----|--------|--------|-----|
|   m |    190 |     90 |   Y |
|   m |    185 |     85 |   Y |
|   m |    180 |     80 |   Y |
|   m |    150 |     80 |   N |
|   m |    180 |     60 |   N |
|   f |    180 |     80 |   N |

You can append the data and train the model. (Typically using AJAX requests.)

# Append the data
$ curl -X POST -F "file=@data.csv" '/rugby?_action=append'

# Re-train the model
$ curl -X POST '/rugby?_action=retrain'
{"score": 1.0}

Now, you can check if a 190cm 80kg boy and a 150cm 80kg boy will get in.

$ curl '/rugby?sex=m&height=190&weight=80'    # 190cm 80kg boy
["Y"]                                         #   will get in
$ curl '/rugby?sex=m&height=150&weight=80'    # 150cm 80kg boy
["N"]                                         #   won't get in

As the users keep adding data and re-training the model, it keeps improving.

Simpler handler names

Handlers are now called micro-services, and identified in gramex.yaml via service: instead of handler:.

The names of these handlers is also simplified. So you can now use:

Going forward, the new version will be preferred, and documentation accordingly. But the old version will continue to work until deprecated.

Other improvements

• Gramex uses openpyxl to read XLSX files, since xlrd has deprecated support for XLSX files. (By Naveen Manukonda)
• gramex <cmd> --help prints help about that command.

Bug fixes

• Redis cache stores have an unlimited default cache size (instead of 500MB). Further, setting size: 0 ignores the cache limit. This is useful in stores like Azure where setting a cache size limit causes an error. (By Mohammed Niyas)
• ProxyHandler no longer fails when gramex.yaml is updated while Gramex is running.

What next

Gramex 1.68 will be released on 1 Feb 2021 and will feature a new “Forms” app. It’s like an embeddable Google Forms that you can use in any app.

Statistics

The Gramex code base has:

• 18,905 lines of Python (645 more than 1.66)
• 1,779 lines of JavaScript (3 more than 1.66)
• 12,045 lines of test code (431 more than 1.66)
• 89% test coverage (same as 1.66)

Credits

• Mohammed Niyas for enabling unlimited-size Redis cache stores
• Naveen Manukonda for always loading XLSX files with openpyxl instead of xlrd

How to install

See the Gramex installation and upgrade instructions

Note: Gramex 1.67 does not work with Python 3.8. We recommend Python 3.7.