A user can play with an Orchestrator base model interactively and improve a snapshot in real-time.
The 'interactive' command starts an interactive session that a user can access an Orchestrator model in real-time doing following:
- Predict the intent of an input utterance using the 'p' commandlet.
- Analyze a model example set, by executing the 'v' (validation) commandlet and produce an evaluation report in real-time.
- Add, remove, or change the intents of an input utterace using the 'a', 'r', and 'c' commandlets, respectively. Users can reference a validation report and choose an ambiguous, misclassified, or low-confidence utterance and change their intent labels.
- Remove some labels completely from the model example set using the 'rl' commandlet.
- Create a new model example set snapshot using the 'n' commandlet.
Below is a list of the commandlets that can be issued during a 'interactive' session Commandlets: h, q, d, s, u, cu, i, ci, ni, cni, q, p, v, vd, va, vm, vl, vat, vlt, vmt, vut, vo, a, r, c, rl, n
- h - print this help message
- q - quit
- d - display utterance, intent label array inputs, Orchestrator config, and the label-index map
- s - show label-utterance statistics of the model examples
- u - enter a new utterance and save it as the "current" utterance input
- cu - clear the "current" utterance input
- i - enter an intent and add it to the "current" intent label array input (can be an index for retrieving a label from the label-index map)
- ci - clear the "current" intent label array input
- ni - enter an intent and add it to the "new" intent label array input (can be an index for retrieving a label from the label-index map)
- cni - clear the "new" intent label array input
- f - find the "current" utterance if it is already in the model example set
- p - make a prediction on the "current" utterance input
- v - validate the model and save analyses (validation report) to "experiment_predicting_va\orchestrator_predicting_set_summary.html"
- vd - reference a validation Duplicates report (previously generated by the "v" command) and enter an index for retrieving utterance/intents into "current"
- va - reference a validation Ambiguous report (previously generated by the "v" command) and enter an index for retrieving utterance/intents into "current"
- vm - reference a validation Misclassified report and enter an index (previously generated by the "v" command) for retrieving utterance/intents into "current"
- vl - reference a validation LowConfidence report (previously generated by the "v" command) and enter an index for retrieving utterance/intents into "current"
- vat - enter a new validation-report ambiguous closeness threshold
- vlt - enter a new validation-report low-confidence threshold
- vmt - enter a new multi-label threshold
- vut - enter a new unknown-label threshold
- vo - enter a boolean for obfuscating labels/utterances or not in evaluation reports generated by the "v" command'
- a - add the "current" utterance and intent labels to the model example set
- r - remove the "current" utterance and intent labels from the model example set
- c - remove the "current" utterance's intent labels and then add it with the "new" intent labels to the model example set
- rl - remove the "current" intent labels from the model example set
- n - create a new snapshot of model examples and save it to "experiment_predicting_va\orchestrator_predicting_training_set.blu"
$ bf orchestrator:interactive --out=resources\data\Columnar\InteractiveOutput --model=resources\data\Columnar\ModelConfig --in=resources\data\Columnar\Email.blu