Usage¶
From the command line¶
deepQuest can be run from the command line by using the commands deepquest or dq followed by either train, predict or score.
Alternatively use -h or --help for information on running deepQuest.
Each mode for deepQuest has some required and optional user input.
Training, Prediction and Scoring¶
1. Training¶
To train a new model, you must pass in a configuration file as a YAML or PKL file using the following syntax for command line or python execution respectively:
dq train -c configs/config-file.yml
or
dq.train('configs/config-file.yml')
dq.train('config.pkl')
For information on configuration files see Tutorial. The pickle option is included for back-compatability of previously trained models and is not applicable to models trained with deepQuest version 2 and above.
Configuration changes can be provided as key=value pairs, over-riding those in the config file, for instance the following commands will resume training of an existing model.
dq train -c configs/config-file.yml --changes RELOAD=True RELOAD_EPOCH=8
dq.train('configs/config-file.yml', {RELOAD:True, RELOAD_EPOCH:8})
Note: The python function takes configuration changes pairs as dictionary elements.
2. Prediction¶
To use a trained model in performing quality estimation inference, use deepQuest’s predict function. Operation is very similar to using train:
dq predict -c configs/config-predict.yml
or
dq.predict('configs/config-predict.yml')
dq.predict('configs/config-file.yml', {RELOAD:True, RELOAD_EPOCH:8})
3. Scoring¶
deepQuest’s scoring function takes two files of the same length, i.e. lists of known and predicted scores, in any order and computes three score metrics: Pearson’s correlation coefficient; Mean Absolute Error; and Root Mean Squared Error.
dq score known.scores predicted.scores
dq.score(['known.scores', 'predicted.scores'])
Where known.scores and predicted.scores are paths to the two files, as a two-element list.