predicted

The predicted-data ensemble as the analyses use it: an (nd, ne) matrix in a layout's row order.

PredictedData

Predictions for every observed cell, one column per member.

Built straight from what each member's simulation returned, so its rows are the layout's rows: the same rows the observation vector and its variance have. The frame the older code passed around is available as a view (:meth:to_frame) for saving and inspection.

nd: int

Number of data rows.

ne: int

Number of members.

from_frame(layout, frame, ne)

From a prediction frame whose cells are (ne,) or (size, ne) arrays.

from_members(layout, members, position=None, scale=None, transform=None)

Fill the matrix from one output per member.

Parameters:

Name	Type	Description	Default
members	sequence	One output per member: a list of records (one dict per report point, keyed by data type) or a DataFrame indexed by label.	required
position	dict	Where each observed label sits in a member's records. Omit when the labels are the positions.	None
scale	(minimum, maximum)	Per-data-type max-min scaling to apply, as the observations were scaled: (value - minimum) / (maximum - minimum).	None
transform	callable	transform(row, values) -> values, applied to a member's (scaled) raw values before they enter the matrix -- how a simulated seismic vintage becomes the wavelet coefficients the observed one was reduced to. Its output must have row.size values; the raw values need not.	None

rows_of(datatype)

The row slices holding datatype, in layout order.

take_members(index)

The predictions of the members index names, in that order.

to_frame(name=None)

The frame view: one cell per observed label and data type.

member_cell(member, row, position)

One member's value for one observed cell, from its records or its frame.