Data passing through Axon layers change type from f64 to f32

Questions / Help
,
1

Sometimes the data passing through Axon layers change type from f64 to f32 . Any workaround for this? (Like enforcing the data type through the pipeline or other workarounds - imagine if it was possible to set the type when calling Explorer.Series.to_tensor(df['col'], type: {:f, 32}))

There is an open issue here.

2

Manged to solve it using Nx.as_type(df, :f32) (but the original problem still remains):

...
  def df_to_tensor(df) do
    df
    |> Explorer.DataFrame.names()
    |> Enum.map(&(df[&1] 
      |> Explorer.Series.to_tensor() 
      |> Nx.new_axis(-1)
      |> Nx.as_type(:f32)))
    |> Nx.concatenate(axis: 1)
  end
...