Skafos

Hi! I'm Skafos
This user has no profile information.
Libraries (View all)

Image classification is the task of assigning a label to an image (classifying an image as one thing or another). Here we start you off with a basic example:

  • Dogs and Cats: Trains a model to classify an image as a cat or a dog using 25,000 images (12,500 Cats and 12,500 Dogs).
  • More Pets: Trains a model to classify an image as a rabbit, mouse, hamster, fish, lizard, or snake.
  • Poison Ivy: Trains a model to classify a plant image as a type of poison ivy (or not).
  • Azure Notebooks: While Notebooks are in mode (beta), some features might be buggy or not available. If you are having trouble with dependencies in your JLab environment. Try waiting a few minutes, restarting the Python 3.6 kernel, and trying again.
  • Training Data: In order for an image classifier to identify a particular category, it must have trained on images labeled as such. For example, if a model was trained on dogs and cats, and it is shown a plant, it will identify that plant as either a dog or cat. To build an image classification model that identifies plants or other types of objects, you would need to retrain the model, using labeled images of the type you want.
  • Model Runtime: The out-of-the box model takes a long time to train on CPU. At this moment, Azure doesn't support GPU training on notebook instances. So if you need that, please check out some of these Google Colab examples.
  • Model Size: In addition to the tips above, try using the model in the function if you are worried about the size of the resulting model. This may also impact the classification accuracy of the model.

Didn't find something you need? Confused by something? Need more guidance?

Please contact us with questions or feedback! Here are two ways:

Also checkout Turi Create's documentation on image classification basics.

Modified on: Jun 18, 2019

Object detection is the task of locating and identifying objects within images. Here we start you off with a basic example:

  • People, Bikes, and Cars: Trains a model to detect and classify objects in images of people, bikes, and cars.
  • Coffee, Pens, and Computers: Trains a model to detect and classify objects in images of coffee mugs, pens and computer monitors.
  • Couches, Chairs, Tables and Beds: Trains a model to detect and classify objects in images of couches, chairs, tables and beds.
  • Azure Notebooks: While Notebooks are in mode (beta), some features might be buggy or not available. If you are having trouble with dependencies in your JLab environment. Try waiting a few minutes, restarting the Python 3.6 kernel, and trying again.
  • Training Data: In order for an object detection model to identify a particular object, it must have seen other objects with the same label. To build an object detection model that identifies what you want, you would need to retrain the model, using bounded and labeled images of the type you want.
  • Model Runtime: The out-of-the box model takes a long time to train on CPU. At this moment, Azure doesn't support GPU training on notebook instances. So if you need that, please check out some of these Google Colab examples.
  • Model Size: Try converting the CoreML model's weights to half-precision if you are worried about the size of the resulting model. This doesn't mean you sacrifice half of your accuracy, it simply means it uses less floating points in the weights of the model. To read more about this, check out Apple's article on this topic.

Didn't find something you need? Confused by something? Need more guidance?

Please contact us with questions or feedback! Here are two ways:

Also check out Turi Create's documentation on object detection basics.

Modified on: Jun 17, 2019

Text classification is the task of assigning a label to some bit of text. Here we start you off with a basic example:

  • Sentiment Classifier: Trains a model to classify user text as positive (5), negative (1), or in between.
  • Spam or Ham: Trains a model to classify user text as "spam" (bad) or "ham" (good).
  • Topic Classifier: Trains a model to classify user text into one of 20 different topics.
  • Azure Notebooks: While Notebooks are in mode (beta), some features might be buggy or not available. If you are having trouble with dependencies in your JLab environment. Try waiting a few minutes, restarting the Python 3.6 kernel, and trying again.
  • Common Text Classification Tasks:
  • Sentiment Classification: How positive or negative is a piece of text? Typically trained with text data representing user reviews and respective ratings on a scale 1-5. The starter model in this repo contains a basic sentiment classifier trained on yelp reviews.
  • Spam Classification: How likely is it that a piece of text is considered "spam" or "ham"? Typically trained with text data paired with the appropriate label (spam or ham). The starter model in this repo contains a basic spam classifier trained on SMS text message data.
  • Topic Identification: What is the topic or subject matter of a piece of text? Typically trained with text data paired with related categories. The starter model in this repo contains a basic topic classifier trained on news articles labeled with corresponding category (provided by scikit-learn package).
  • Wrangling Text w/ Turi Create: Text comes in a host of different formats. Fortunately, the Turi Create text classifier handles all tokenization, and text feature engineering, and cleaning of your text data automatically.

Didn't find something you need? Confused by something? Need more guidance?

Please contact us with questions or feedback! Here are two ways:

Also checkout Turi Create's documentation on text classification basics.

Modified on: Jun 17, 2019
Starred Libraries (View all starred libraries)