What's currently supported?
This article will help you understand what type of problems and machine learning frameworks are supported by the Quality Analysis.
Frameworks
The supported frameworks are Tensorflow's (Keras) sequential and model API; and Scikit-Learn's gradient boosting, logistic regression, and random forest (Classifier and Regressor).
Input data shape
Snitch currently supports tabular datasets (2D) for the training set and column (1D) or tabular (2D) datasets for the output dataset. In other words, Snitch supports classical machine learning classification or regression problems with single or multi-output.
Saving methods
The supported saving methods are:
Library | Code | Extension | Link to documentation | |
Keras/Tensorflow | model.save(file_path) | .h5 | Tensorflow: Save and Serialize | |
Scikit-Learn | joblib.dump(model, file_path) pickle.dump(model, file_path) |
.joblib or .pkl | Scikit-Learn: Model Persistence |
String/categorical features
Models expecting data or targets to string are not supported. You should encode categorical features and targets as numerical values before training your model if you eventually want to perform a Quality Analysis. Preprocessing techniques like OneHotEncoding can be used to do so.
Classification targets
For classification problems, either class (0,1,2,...) or logits ( (1,0,0), (0,1,0), (0,0,1) ) as targets are accepted.
Time series
Time series are supported as long as they are formatted into tabular datasets (i.e., not sequential or 3D). Datetime values are however not supported. See this article for more information on how to format your dataset to use Snitch.
Natural language processing (NLP) and computer vision
Snitch does not support Natural Language Processing (NLP) and Computer Vision models.