Basic bits#

from qtypy import dataflow
df = dataflow(multithreaded = True)

Computing columns#

A column can be computed in 4 ways:

  • dataset.column whose value is read in from an existing dataset.

  • column.constant whose value does not change throughout per-event.

  • column.expression, or simply a str, that is jitted using Cling.

  • column.definition, a user-compiled C++ computation unit (see API reference for more details).

from qtypy import dataset, column

df["x"] = dataset.column(key="x", dtype="ROOT::RVec<float>")
df["y"] = "x.size()"

df["invalid C++ identifier"] = column.constant(1)  # ❌

Applying selections#

By default, successive selection operations are always compounded one after another:

(
df
.filter({"all" : "true"})       # cut = true,       weight = 1.0
.filter({"a"   : "a == true"})  # cut = (all && a), weight = 1.0
.weight({"w"   : "w == true"})  # cut = (all && a), weight = (1.0 * w)
)

Branching & merging#

Diverging selections can be applied by specifying they be applied “at” a branching point:

a = df & {"a" : "a == true"}
a_and_b = df @ "a" & {"a_and_b" : "b == true"}
a_and_b = df @ "a" & {"a_and_b" : "c == true"}

Divergent selections can be subsequently re-merged similarly:

intersection = (df @ "a") & {"intersection" : "a_and_b && a_and_c"}
union        = (df @ "a") & {"union"  : "a_and_b || a_and_c"}

Histogram outputs#

from qtypy import hist

# from whatever the last applied selection was
hx = df >> hist("x", dtype="float", nx=100, xmin=0.0, xmax=1.0).fill("x")

df @ "base" >> hist("x", dtype="float", nx=100, xmin=0.0, xmax=1.0).fill("x")

Remember that data processing is triggered only once the result of a query is explicitly accessed:

hx.result()  # TH1F

Multiple queries can be performed at once (of course). Query definitions can be “recycled” to be run at multiple selections, for example:

# multiple selections
selections = ["a", "b", "c"]

# define query without df
hx = hist("x", dtype="float", nx=100, xmin=0.0, xmax=1.0).fill("x")

# query at multiple selections
hxs = {
    sel : df @ sel >> hx for sel in selections
}

Important

Accessing the result of a query triggers the event-loop (if needed). In order to prevent running it multiple times, make sure that you have specified all queries needed at a time before accessing any of their results!

Numpy array outputs#

x = df >> column.to_numpy("x", dtype="float")
x.result()  # <class "numpy.ndarray">, dtype("float32")

Note

numpy.asarray() supports zero-copy readout of the std::vector only in the case of scalar elements. For nested/arbitrary data types, a copy must be invoked when crossing the C++-to-Python boundary.