axelfahy / bff Goto Github PK

Currently, the floating precision is hard-coded as followed:

ax.yaxis.set_major_formatter(
    mpl.ticker.FuncFormatter(lambda x, p: f'{x:,.1f}')
)

Change to detect automatically the number of decimal needed on the plot.

Add function to automatically cast the column to categorical type if it is memory efficient.

Add some images of the outpout of the plots in the documentation and a notebook with examples.

Add a function to plot boxplot in a pretty way.

Add function to pretty plot correlation between variables.

Need to build documentation and put it online.

• Apply the same convention for the disposition of the arguments (e.g. one by row) and check if the order is logical for plot's arguments.

• Use Optional type instead of Union[t, None].

The constructor of the FancyConfig class only accept Path object. It should accept strings, test the type and create the Path object in case of a string. The rest of the process stays the same.

Currently, arguments of plot functions are redundant between functions. Parameters such as figsize, dpi, etc. are common to all functions.

Moreover, some style, as alpha, fontsize, are common between the functions.

Find a way to set those parameters without repeating them.

Add parameter rot for plot functions to rotate the x label with the given angle.

X axis is the frequencies and Y axis the power. Should probably be implemented using scipy.signal.periodogram or scipy.signal.welch.

Add a flag to plot the distribution on the side of the plot.

Pandas related functions should be put inside a pd module.
Remaining functions should go inside a core module.
This will break the current api and should be release for the 0.3 version of the package.

Suffix _pd of pandas related function might be redundant and might be removed as well. It is to be considered.

Currently, the metric is plotted in another subplot, below the loss of the history in the plot_history function.
Add an option to have both information on the same subplot, using twinx.
Use four colors for the plot instead of the main two.

Separation of plots in bff.plot module.

Add a minimal coverage for the validation when building.

Default location for loc in legend is best, which raise the following warning when used with lots of data: UserWarning: Creating legend with loc="best" can be slow with large amounts of data.

Add the possibility to specify the location of the legend with the parameters function.

Non-exhaustive list of useful logging methods that could be used in this context:

(df
 .assign(foo=df['bar'] ** 2) 
 .pipe(log_dtypes)
 .to_csv('foo_file.csv')
)

• log_dtypes(df: pd.DataFrame)
• log_shape(df: pd.DataFrame)
• log_columns(df: pd.DataFrame)
• log_msg(df: pd.DataFrame, msg: str, fun: Callable = print, **kwargs) (can be given print, or logging method, or any other callable, any output)
• ...