MultipleSeries to dataframe

[CommonClimate]

Generalizing the creation of Series objects from a dataFrame would be very helpful. See Dry_Tortugas.ipynb for an example: currently we create different Series manually (one per dataframe column), but it would be very helpful to have a pyleo.MultipleSeries methods called from_pandas() that would take the following arguments:

• df, a Pandas dataframe
• time_col, name or index of the time used to create the time variable, hence the datetime index
• value_cols: index of the columns to be exported into a MultipleSeries object (typically, users will only want certain columns from a spreadsheet)

metadata can be inferred from the spreadsheet columns (e.g. a string in parentheses is likely to be the value_unit of the corresponding Series, like "mmol/mol" in this example; the string preceding that parenthesis ("Sr/Ca" in this example) will be the value_name of the corresponding Series, or the name of the pd.Series created from that column of the DataFrame. Similarly, metadata for the time variable (particularly the all-important time_unit) can be inferred from the header of df.iloc[:,time_col].

The method should return a MultipleSeries object (i.e. a list of pyleo.Series objects).

[CommonClimate]

I take Deborah's point that this will be impossible to do for general cases, because each paleo spreadsheet is a bit of a snowflake. However, the reverse is possible: pyleo.MultipleSeries objects should have a to_pandas() method just like pyleo.Series objects do. Instead of exporting to pd.Series, it would export to pd.DataFrame. This would likely require some kind of alignment to make place, therefore issue #28 takes precedence.

[MarcoGorelli]

So MultipleSeries.from_pandas is no longer a requirement?

MultipleSeries.to_pandas - doesn't look like this requires #28 to be done first, given that you already have a way of aligning Series

[CommonClimate]

Addressed by LinkedEarth/Pyleoclim_util#335

[CommonClimate]

Question re: this. As shown in Paleopandas Playground.ipynb, I get a result like this in my test case:

this is too good to be true: all three Series in this object have completely different time axes, so at first I could not understand how the df appears to have one common index to them all and no missing values (NaNs). Looking into the function, I saw that it applies common_time() under the hood, but I think this should be an optional parameter. The default behavior should be to align the timeseries to the same index, but keep NaNs in the DataFrame, unless you see a major objection to that. At the very least, users need to be warned that common_time() has been applied with default parameters. I can actually do that by applying the concept of log (present in Series) to this class as well.

[CommonClimate]

addressed by #350