Dagli provides a number of well-commented examples in the dagli/examples/ directory. These are
generally toy problems but can be a useful guide when getting started with Dagli.
Located in dagli/examples/fasttext-and-avro,
this example reads dialog from Shakespearean plays from an Avro record using a @Struct, and then learns a FastText model
to predict a character given their dialog (e.g. "A rose by any other name would smell as sweet" -> "Juliet").
- FastText text classification model
- Reading Avro data with @Structs
- Evaluating, serializing, deserializing and applying the trained model
Located in dagli/examples/complex-model,
this example reads dialog from Shakespearean plays from a tab-separated value file and also learns to predict a
character given their dialog.
- Pipelined text classification model employing both FastText and XGBoost
- Creating and combining feature vectors
- Using
BestModelfor model selection - Reading delimited-separated values (DSV) with
DSVReader - Evaluating, serializing, deserializing and applying the trained model
Located in dagli/examples/neural-network,
several types of neural network classification models are demonstrated:
- A bag-of-ngrams feed-forward model
- A pseudo-Transformer-encoder model using multi-headed attention
- A letter-by-letter LSTM (RNN) model
Located in dagli/examples/xgboost-and-liblinear,
this example demonstrates a common pipelined model in industry: training an XGBoost classifier and then using the active
leaves for each example as features in a downstream logistic regressor (together with many other features).
Located in dagli/examples/assorted, this is a
collection of smaller and simpler examples:
- Definition of new, prepared transformers
- A pipelined model that uses FastText embeddings in a logistic regression classifier
- A logistic regression model using Word2Vec pretrained embeddings
- A straightforward XGBoost model