Using ColumnarModelData.from_data_frame I am always getting accuracy of 96%
Recreating Rossman worked fine but with a different dataset I have problems.
Is there something obvious I am missing? I suspect that since this data set is looking for a % instead of sales number there is something strange going on.
We have tried altering:
- Learning Rate
- Batch Size
- Embeddings
- Layer sizes
- y_Range
- Target size ( x100 so that it is not 0 and 1 in training, hence the screenshot above)
- Replacing negative numbers with 0s