Fix HistoryLogger for training with epochs=0

octoenergy / timeserio

Better `keras` models for time series and beyond

MIT License

61 stars 16 forks source link

Fix HistoryLogger for training with epochs=0 #33

Closed ig248 closed 3 years ago

ig248 commented 3 years ago

Before this change, training with epochs=0 (valid in keras and useful for getting a loss baseline/debugging) would lead to errors in the HistoryLogger.

After this change, suitable defaults are substituted.