I noticed that the Fallback don’t really works by me.
When I type in some random letters like “sdfasdf” or “blablabla” it recognise them as names with a confidence around 0.91
Is this because of my configurations or because the nul_fallback has only 1 training examples?
(or are this common names anywhere in the world?
)
my config.yml looks like this:
language: de_core_news_md
pipeline:
- name: SpacyNLP
- name: SpacyTokenizer
- name: SpacyFeaturizer
- name: RegexFeaturizer
- name: LexicalSyntacticFeaturizer
- name: CountVectorsFeaturizer
- name: CountVectorsFeaturizer
analyzer: char_wb
min_ngram: 1
max_ngram: 4
- name: DIETClassifier
epochs: 300
- name: EntitySynonymMapper
- name: ResponseSelector
epochs: 100
- name: FallbackClassifier
threshold: 0.3
ambiguity_threshold: 0.1
policies:
- name: MemoizationPolicy
- name: TEDPolicy
max_history: 12
epochs: 100
- name: RulePolicy
core_fallback_threshold: 0.3
core_fallback_action_name: "action_default_fallback"