26#include <nlohmann/json.hpp>
1500 void Exec(unsigned int slot)
1502 fPerThreadResults[slot]++;
1505 // Called at the end of the event loop.
1508 *fFinalResult = std::accumulate(fPerThreadResults.begin(), fPerThreadResults.end(), 0);
1511 // Called by RDataFrame to retrieve the name of this action.
1512 std::string GetActionName() const { return "MyCounter"; }
1516 ROOT::RDataFrame df(10);
1517 ROOT::RDF::RResultPtr<int> resultPtr = df.Book<>(MyCounter{df.GetNSlots()}, {});
1518 // The GetValue call triggers the event loop
1519 std::cout << "Number of processed entries: " << resultPtr.GetValue() << std::endl;
1523See the Book() method for more information and [this tutorial](https://root.cern/doc/master/df018__customActions_8C.html)
1524for a more complete example.
1526#### Injecting arbitrary code in the event loop with Foreach() and ForeachSlot()
1528Foreach() takes a callable (lambda expression, free function, functor...) and a list of columns and
1529executes the callable on the values of those columns for each event that passes all upstream selections.
1530It can be used to perform actions that are not already available in the interface. For example, the following snippet
1531evaluates the root mean square of column "x":
1533// Single-thread evaluation of RMS of column "x" using Foreach
1536df.Foreach([&sumSq, &n](double x) { ++n; sumSq += x*x; }, {"x"});
1537std::cout << "rms of x: " << std::sqrt(sumSq / n) << std::endl;
1539In multi-thread runs, users are responsible for the thread-safety of the expression passed to Foreach():
1540thread will execute the expression concurrently.
1541The code above would need to employ some resource protection mechanism to ensure non-concurrent writing of `rms`; but
1542this is probably too much head-scratch for such a simple operation.
1544ForeachSlot() can help in this situation. It is an alternative version of Foreach() for which the function takes an
1545additional "processing slot" parameter besides the columns it should be applied to. RDataFrame
1546guarantees that ForeachSlot() will invoke the user expression with different `slot` parameters for different concurrent
1547executions (see [Special helper columns: rdfentry_ and rdfslot_](\ref helper-cols) for more information on the slot parameter).
1548We can take advantage of ForeachSlot() to evaluate a thread-safe root mean square of column "x":
1550// Thread-safe evaluation of RMS of column "x" using ForeachSlot
1551ROOT::EnableImplicitMT();
1552const unsigned int nSlots = df.GetNSlots();
1553std::vector<double> sumSqs(nSlots, 0.);
1554std::vector<unsigned int> ns(nSlots, 0);
1556df.ForeachSlot([&sumSqs, &ns](unsigned int slot, double x) { sumSqs[slot] += x*x; ns[slot] += 1; }, {"x"});
1557double sumSq = std::accumulate(sumSqs.begin(), sumSqs.end(), 0.); // sum all squares
1558unsigned int n = std::accumulate(ns.begin(), ns.end(), 0); // sum all counts
1559std::cout << "rms of x: " << std::sqrt(sumSq / n) << std::endl;
1561Notice how we created one `double` variable for each processing slot and later merged their results via `std::accumulate`.
1565### Dataset joins with friend trees
1567Vertically concatenating multiple trees that have the same columns (creating a logical dataset with the same columns and
1568more rows) is trivial in RDataFrame: just pass the tree name and a list of file names to RDataFrame's constructor, or create a TChain
1569out of the desired trees and pass that to RDataFrame.
1571Horizontal concatenations of trees or chains (creating a logical dataset with the same number of rows and the union of the
1572columns of multiple trees) leverages TTree's "friend" mechanism.
1574Simple joins of trees that do not have the same number of rows are also possible with indexed friend trees (see below).
1576To use friend trees in RDataFrame, set up trees with the appropriate relationships and then instantiate an RDataFrame
1582main.AddFriend(&friend, "myFriend");
1585auto df2 = df.Filter("myFriend.MyCol == 42");
1588The same applies for TChains. Columns coming from the friend trees can be referred to by their full name, like in the example above,
1589or the friend tree name can be omitted in case the column name is not ambiguous (e.g. "MyCol" could be used instead of
1590"myFriend.MyCol" in the example above if there is no column "MyCol" in the main tree).
1592\note A common source of confusion is that trees that are written out from a multi-thread Snapshot() call will have their
1593 entries (block-wise) shuffled with respect to the original tree. Such trees cannot be used as friends of the original
1594 one: rows will be mismatched.
1596Indexed friend trees provide a way to perform simple joins of multiple trees over a common column.
1597When a certain entry in the main tree (or chain) is loaded, the friend trees (or chains) will then load an entry where the
1598"index" columns have a value identical to the one in the main one. For example, in Python:
1604# If a friend tree has an index on `commonColumn`, when the main tree loads
1605# a given row, it also loads the row of the friend tree that has the same
1606# value of `commonColumn`
1607aux_tree.BuildIndex("commonColumn")
1609mainTree.AddFriend(aux_tree)
1611df = ROOT.RDataFrame(mainTree)
1614RDataFrame supports indexed friend TTrees from ROOT v6.24 in single-thread mode and from v6.28/02 in multi-thread mode.
1616\anchor other-file-formats
1617### Reading data formats other than ROOT trees
1618RDataFrame can be interfaced with RDataSources. The ROOT::RDF::RDataSource interface defines an API that RDataFrame can use to read arbitrary columnar data formats.
1620RDataFrame calls into concrete RDataSource implementations to retrieve information about the data, retrieve (thread-local) readers or "cursors" for selected columns
1621and to advance the readers to the desired data entry.
1622Some predefined RDataSources are natively provided by ROOT such as the ROOT::RDF::RCsvDS which allows to read comma separated files:
1624auto tdf = ROOT::RDF::FromCSV("MuRun2010B.csv");
1625auto filteredEvents =
1626 tdf.Filter("Q1 * Q2 == -1")
1627 .Define("m", "sqrt(pow(E1 + E2, 2) - (pow(px1 + px2, 2) + pow(py1 + py2, 2) + pow(pz1 + pz2, 2)))");
1628auto h = filteredEvents.Histo1D("m");
1632See also FromNumpy (Python-only), FromRNTuple(), FromArrow(), FromSqlite().
1635### Computation graphs (storing and reusing sets of transformations)
1637As we saw, transformed dataframes can be stored as variables and reused multiple times to create modified versions of the dataset. This implicitly defines a **computation graph** in which
1638several paths of filtering/creation of columns are executed simultaneously, and finally aggregated results are produced.
1640RDataFrame detects when several actions use the same filter or the same defined column, and **only evaluates each
1641filter or defined column once per event**, regardless of how many times that result is used down the computation graph.
1642Objects read from each column are **built once and never copied**, for maximum efficiency.
1643When "upstream" filters are not passed, subsequent filters, temporary column expressions and actions are not evaluated,
1644so it might be advisable to put the strictest filters first in the graph.
1646\anchor representgraph
1647### Visualizing the computation graph
1648It is possible to print the computation graph from any node to obtain a [DOT (graphviz)](https://en.wikipedia.org/wiki/DOT_(graph_description_language)) representation either on the standard output
1651Invoking the function ROOT::RDF::SaveGraph() on any node that is not the head node, the computation graph of the branch
1652the node belongs to is printed. By using the head node, the entire computation graph is printed.
1654Following there is an example of usage:
1656// First, a sample computational graph is built
1657ROOT::RDataFrame df("tree", "f.root");
1659auto df2 = df.Define("x", []() { return 1; })
1660 .Filter("col0 % 1 == col0")
1661 .Filter([](int b1) { return b1 <2; }, {"cut1"})
1662 .Define("y", []() { return 1; });
1664auto count = df2.Count();
1666// Prints the graph to the rd1.dot file in the current directory
1667ROOT::RDF::SaveGraph(df, "./mydot.dot");
1668// Prints the graph to standard output
1669ROOT::RDF::SaveGraph(df);
1672The generated graph can be rendered using one of the graphviz filters, e.g. `dot`. For instance, the image below can be generated with the following command:
1674$ dot -Tpng computation_graph.dot -ocomputation_graph.png
1677\image html RDF_Graph2.png
1680### Activating RDataFrame execution logs
1682RDataFrame has experimental support for verbose logging of the event loop runtimes and other interesting related information. It is activated as follows:
1684#include <ROOT/RLogger.hxx>
1686// this increases RDF's verbosity level as long as the `verbosity` variable is in scope
1687auto verbosity = ROOT::RLogScopedVerbosity(ROOT::Detail::RDF::RDFLogChannel(), ROOT::ELogLevel::kInfo);
1694verbosity = ROOT.RLogScopedVerbosity(ROOT.Detail.RDF.RDFLogChannel(), ROOT.ELogLevel.kInfo)
1697More information (e.g. start and end of each multi-thread task) is printed using `ELogLevel.kDebug` and even more
1698(e.g. a full dump of the generated code that RDataFrame just-in-time-compiles) using `ELogLevel.kDebug+10`.
1700\anchor rdf-from-spec
1701### Creating an RDataFrame from a dataset specification file
1703RDataFrame can be created using a dataset specification JSON file:
1708df = ROOT.RDF.Experimental.FromSpec("spec.json")
1711The input dataset specification JSON file needs to be provided by the user and it describes all necessary samples and
1712their associated metadata information. The main required key is the "samples" (at least one sample is needed) and the
1713required sub-keys for each sample are "trees" and "files". Additionally, one can specify a metadata dictionary for each
1714sample in the "metadata" key.
1716A simple example for the formatting of the specification in the JSON file is the following:
1722 "trees": ["tree1", "tree2"],
1723 "files": ["file1.root", "file2.root"],
1727 "sample_category" = "data"
1731 "trees": ["tree3", "tree4"],
1732 "files": ["file3.root", "file4.root"],
1736 "sample_category" = "MC_background"
1743The metadata information from the specification file can be then accessed using the DefinePerSample function.
1744For example, to access luminosity information (stored as a double):
1747df.DefinePerSample("lumi", 'rdfsampleinfo_.GetD("lumi")')
1750or sample_category information (stored as a string):
1753df.DefinePerSample("sample_category", 'rdfsampleinfo_.GetS("sample_category")')
1756or directly the filename:
1759df.DefinePerSample("name", "rdfsampleinfo_.GetSampleName()")
1762An example implementation of the "FromSpec" method is available in tutorial: df106_HiggstoFourLeptons.py, which also
1763provides a corresponding exemplary JSON file for the dataset specification.
1766### Adding a progress bar
1768A progress bar showing the processed event statistics can be added to any RDataFrame program.
1769The event statistics include elapsed time, currently processed file, currently processed events, the rate of event processing
1770and an estimated remaining time (per file being processed). It is recorded and printed in the terminal every m events and every
1771n seconds (by default m = 1000 and n = 1). The ProgressBar can be also added when the multithread (MT) mode is enabled.
1773ProgressBar is added after creating the dataframe object (df):
1775ROOT::RDataFrame df("tree", "file.root");
1776ROOT::RDF::Experimental::AddProgressBar(df);
1779Alternatively, RDataFrame can be cast to an RNode first, giving the user more flexibility
1780For example, it can be called at any computational node, such as Filter or Define, not only the head node,
1781with no change to the ProgressBar function itself (please see the [Python interface](classROOT_1_1RDataFrame.html#python)
1782section for appropriate usage in Python):
1784ROOT::RDataFrame df("tree", "file.root");
1785auto df_1 = ROOT::RDF::RNode(df.Filter("x>1"));
1786ROOT::RDF::Experimental::AddProgressBar(df_1);
1788Examples of implemented progress bars can be seen by running [Higgs to Four Lepton tutorial](https://root.cern/doc/master/df106__HiggsToFourLeptons_8py_source.html) and [Dimuon tutorial](https://root.cern/doc/master/df102__NanoAODDimuonAnalysis_8C.html).
1790\anchor missing-values
1791### Working with missing values in the dataset
1793In certain situations a dataset might be missing one or more values at one or
1794more of its entries. For example:
1796- If the dataset is composed of multiple files and one or more files is
1797 missing one or more columns required by the analysis.
1798- When joining different datasets horizontally according to some index value
1799 (e.g. the event number), if the index does not find a match in one or more
1800 other datasets for a certain entry.
1801- If, for a certain event, a column is invalid because it results from a Snapshot
1802 with systematic variations, and that variation didn't pass its filters. For
1803 more details, see \ref snapshot-with-variations.
1805For example, suppose that column "y" does not have a value for entry 42:
1815If the RDataFrame application reads that column, for example if a Take() action
1816was requested, the default behaviour is to throw an exception indicating
1817that that column is missing an entry.
1819The following paragraphs discuss the functionalities provided by RDataFrame to
1820work with missing values in the dataset.
1822#### FilterAvailable and FilterMissing
1824FilterAvailable and FilterMissing are specialized RDataFrame Filter operations.
1825They take as input argument the name of a column of the dataset to watch for
1826missing values. Like Filter, they will either keep or discard an entire entry
1827based on whether a condition returns true or false. Specifically:
1829- FilterAvailable: the condition is whether the value of the column is present.
1830 If so, the entry is kept. Otherwise if the value is missing the entry is
1832- FilterMissing: the condition is whether the value of the column is missing. If
1833 so, the entry is kept. Otherwise if the value is present the entry is
1837df = ROOT.RDataFrame(dataset)
1839# Anytime an entry from "col" is missing, the entire entry will be filtered out
1840df_available = df.FilterAvailable("col")
1841df_available = df_available.Define("twice", "col * 2")
1843# Conversely, if we want to select the entries for which the column has missing
1844# values, we do the following
1845df_missingcol = df.FilterMissing("col")
1846# Following operations in the same branch of the computation graph clearly
1847# cannot access that same column, since there would be no value to read
1848df_missingcol = df_missingcol.Define("observable", "othercolumn * 2")
1852ROOT::RDataFrame df{dataset};
1854// Anytime an entry from "col" is missing, the entire entry will be filtered out
1855auto df_available = df.FilterAvailable("col");
1856auto df_twicecol = df_available.Define("twice", "col * 2");
1858// Conversely, if we want to select the entries for which the column has missing
1859// values, we do the following
1860auto df_missingcol = df.FilterMissing("col");
1861// Following operations in the same branch of the computation graph clearly
1862// cannot access that same column, since there would be no value to read
1863auto df_observable = df_missingcol.Define("observable", "othercolumn * 2");
1868DefaultValueFor creates a node of the computation graph which just forwards the
1869values of the columns necessary for other downstream nodes, when they are
1870available. In case a value of the input column passed to this function is not
1871available, the node will provide the default value passed to this function call
1875df = ROOT.RDataFrame(dataset)
1876# Anytime an entry from "col" is missing, the value will be the default one
1877default_value = ... # Some sensible default value here
1878df = df.DefaultValueFor("col", default_value)
1879df = df.Define("twice", "col * 2")
1883ROOT::RDataFrame df{dataset};
1884// Anytime an entry from "col" is missing, the value will be the default one
1885constexpr auto default_value = ... // Some sensible default value here
1886auto df_default = df.DefaultValueFor("col", default_value);
1887auto df_col = df_default.Define("twice", "col * 2");
1890#### Mixing different strategies to work with missing values in the same RDataFrame
1892All the operations presented above only act on the particular branch of the
1893computation graph where they are called, so that different results can be
1894obtained by mixing and matching the filtering or providing a default value
1898df = ROOT.RDataFrame(dataset)
1899# Anytime an entry from "col" is missing, the value will be the default one
1900default_value = ... # Some sensible default value here
1901df_default = df.DefaultValueFor("col", default_value).Define("twice", "col * 2")
1902df_filtered = df.FilterAvailable("col").Define("twice", "col * 2")
1904# Same number of total entries as the input dataset, with defaulted values
1905df_default.Display(["twice"]).Print()
1906# Only keep the entries where "col" has values
1907df_filtered.Display(["twice"]).Print()
1911ROOT::RDataFrame df{dataset};
1913// Anytime an entry from "col" is missing, the value will be the default one
1914constexpr auto default_value = ... // Some sensible default value here
1915auto df_default = df.DefaultValueFor("col", default_value).Define("twice", "col * 2");
1916auto df_filtered = df.FilterAvailable("col").Define("twice", "col * 2");
1918// Same number of total entries as the input dataset, with defaulted values
1919df_default.Display({"twice"})->Print();
1920// Only keep the entries where "col" has values
1921df_filtered.Display({"twice"})->Print();
1924#### Further considerations
1926Note that working with missing values is currently supported with a TTree-based
1927data source. Support of this functionality for other data sources may come in
1930\anchor special-values
1931### Dealing with NaN or Inf values in the dataset
1933RDataFrame does not treat NaNs or infinities beyond what the floating-point standards require, i.e. they will
1934propagate to the final result.
1935Non-finite numbers can be suppressed using Filter(), e.g.:
1938df.Filter("std::isfinite(x)").Mean("x")
2073namespace Experimental {
2127 auto *
lm = df->GetLoopManager();
2129 throw std::runtime_error(
"Cannot print information about this RDataFrame, "
2130 "it was not properly created. It must be discarded.");
2132 auto defCols =
lm->GetDefaultColumnNames();
2134 std::ostringstream
ret;
2135 if (
auto ds = df->GetDataSource()) {
2136 ret <<
"A data frame associated to the data source \"" << cling::printValue(
ds) <<
"\"";
2138 ret <<
"An empty data frame that will create " <<
lm->GetNEmptyEntries() <<
" entries\n";
Basic types used by ROOT and required by TInterpreter.
unsigned long long ULong64_t
Portable unsigned long integer 8 bytes.
ROOT::Detail::TRangeCast< T, true > TRangeDynCast
TRangeDynCast is an adapter class that allows the typed iteration through a TCollection.
The head node of a RDF computation graph.
The dataset specification for RDataFrame.
std::shared_ptr< ROOT::Detail::RDF::RLoopManager > fLoopManager
< The RLoopManager at the root of this computation graph. Never null.
ROOT's RDataFrame offers a modern, high-level interface for analysis of data stored in TTree ,...
RDataFrame(std::string_view treeName, std::string_view filenameglob, const ColumnNames_t &defaultColumns={})
Build the dataframe.
ROOT::RDF::ColumnNames_t ColumnNames_t
Describe directory structure in memory.
A TTree represents a columnar dataset.
ROOT::RDF::Experimental::RDatasetSpec RetrieveSpecFromJson(const std::string &jsonFile)
Function to retrieve RDatasetSpec from JSON file provided.
ROOT::RDataFrame FromSpec(const std::string &jsonFile)
Factory method to create an RDataFrame from a JSON specification file.
std::vector< std::string > ColumnNames_t
std::shared_ptr< const ColumnNames_t > ColumnNamesPtr_t
