Loss value changes without any changes in training data or configuration in RASA NLU

Question

I am working with RASA NLU to extract intents, and i faced a problem which is loss value changes every single time i train the model even when no new training data and no edits in config. And these changes affect badly my results to extract intents.

I use this configuration:

language: "ar" pipeline: - name: "tokenizer_whitespace" - name: "ner_crf" - name: "ner_synonyms" - name: "intent_featurizer_count_vectors" - name: "intent_classifier_tensorflow_embedding"

How to avoid such a problem, and keep the model consistent?

Answer 1

I got the answer, you need to set the "random_seed" parameter under "tensorflow_embedding", to a fixed number in your configurations file.