I was trying to set up my pipeline to output tensorboard logs, however I noticed in the graphs that the validation metrics stop after a couple of steps, whereas the training metrics continue
I can also see it happening in the terminal, where at some point all the
val_ metrics disappear.
Here is my config:
language: en pipeline: - name: WhitespaceTokenizer - name: RegexFeaturizer - name: LexicalSyntacticFeaturizer - name: CountVectorsFeaturizer - name: CountVectorsFeaturizer analyzer: char_wb min_ngram: 1 max_ngram: 4 - name: DIETClassifier batch_size: 32 epochs: 20 constrain_similarities: true evaluate_on_number_of_examples: 880 # 15% of the data for validation evaluate_every_number_of_epochs: 2 tensorboard_log_directory: "models/tensorboard/v4" tensorboard_log_level: "epoch" - name: EntitySynonymMapper - name: FallbackClassifier threshold: 0.3 - name: ResponseSelector epochs: 5 evaluate_every_number_of_epochs: 1 evaluate_on_number_of_examples: 20 constrain_similarities: true policies: - name: MemoizationPolicy - name: RulePolicy core_fallback_threshold: 0.6 core_fallback_action_name: action_default_fallback enable_fallback_prediction: true - name: TEDPolicy max_history: 7 epochs: 1 constrain_similarities: true