Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Tau L1 2023 #1087

Merged
merged 7 commits into from
Aug 30, 2023
Merged

Tau L1 2023 #1087

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
9fc02fa
Add retrained L1 Tau, HW option, and Tau GT Format
Duchstf Mar 26, 2023
f301913
fix for GT format tau eta and phi
Duchstf Jun 1, 2023
509ce67
remove more unnecessary codes
Duchstf Jun 7, 2023
ac63ddd
remove eta and phi
Duchstf Jun 7, 2023
7d1408d
add comments and const
Duchstf Jun 21, 2023
23e7756
pull changes
Duchstf Jun 28, 2023
dd376b2
code format and successfully compiled
Duchstf Aug 24, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 17 additions & 6 deletions DataFormats/L1TParticleFlow/interface/PFTau.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,14 +4,16 @@
#include <algorithm>
#include <vector>
#include "DataFormats/L1Trigger/interface/L1Candidate.h"
#include "DataFormats/L1TParticleFlow/interface/taus.h"
#include "DataFormats/L1TParticleFlow/interface/gt_datatypes.h"

namespace l1t {

static constexpr float PFTAU_NN_OFFSET = 0.1;
static constexpr float PFTAU_NN_SLOPE = 0.2;
static constexpr float PFTAU_NN_OVERALL_SCALE = 1. / 20.1;

static constexpr float PFTAU_NN_LOOSE_CUT = 0.05;
static constexpr float PFTAU_NN_LOOSE_CUT = 0.28;
static constexpr float PFTAU_NN_TIGHT_CUT = 0.25;

static constexpr float PFTAU_PF_LOOSE_CUT = 10.0;
Expand All @@ -26,14 +28,16 @@ namespace l1t {
PFTau() {}
enum { unidentified = 0, oneprong = 1, oneprongpi0 = 2, threeprong = 3 };
PFTau(const LorentzVector& p,
float iVector[80],
Duchstf marked this conversation as resolved.
Show resolved Hide resolved
float iso = -1,
float fulliso = -1,
int id = 0,
int hwpt = 0,
int hweta = 0,
int hwphi = 0)
: PFTau(PolarLorentzVector(p), iso, id, hwpt, hweta, hwphi) {}
: PFTau(PolarLorentzVector(p), iVector, iso, id, hwpt, hweta, hwphi) {}
PFTau(const PolarLorentzVector& p,
float iVector[80],
float iso = -1,
float fulliso = -1,
int id = 0,
Expand All @@ -49,12 +53,10 @@ namespace l1t {

float z0() const { return vz(); }
float dxy() const { return dxy_; }
const float* NNValues() const { return NNValues_; }

bool passMass() const { return (mass() < 2 + pt() / PTSCALING_MASSCUT); }
bool passLooseNN() const {
return iso_ * (PFTAU_NN_OFFSET + PFTAU_NN_SLOPE * (min(pt(), PFTAU_NN_PT_CUTOFF))) * PFTAU_NN_OVERALL_SCALE >
PFTAU_NN_LOOSE_CUT;
}
bool passLooseNN() const { return iso_ > PFTAU_NN_LOOSE_CUT; }
bool passLooseNNMass() const {
if (!passMass())
return false;
Expand All @@ -72,11 +74,20 @@ namespace l1t {
}
bool passTightPF() const { return fullIso_ < PFTAU_PF_TIGHT_CUT; }

//Tau encoding for GT
void set_encodedTau(l1gt::PackedTau encodedTau) { encodedTau_ = encodedTau; }
l1gt::PackedTau encodedTau() const { return encodedTau_; } //Can be unpacked using l1gt::Tau::unpack()

//Return the l1gt Tau object from the encoded objects
l1gt::Tau getHWTauGT() const { return l1gt::Tau::unpack(encodedTau_); }

private:
float NNValues_[80]; // Values for each of the 80 NN inputs
float iso_;
float fullIso_;
int id_;
float dxy_;
l1gt::PackedTau encodedTau_;
};

typedef std::vector<l1t::PFTau> PFTauCollection;
Expand Down
44 changes: 42 additions & 2 deletions DataFormats/L1TParticleFlow/interface/gt_datatypes.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,8 @@ namespace l1gt {

// tau fields
typedef ap_ufixed<10, 8> tauseed_pt_t;
typedef ap_uint<10> tau_rawid_t;
typedef std::array<uint64_t, 2> PackedTau;

namespace Scales {
const int INTPHI_PI = 1 << (phi_t::width - 1);
Expand Down Expand Up @@ -185,12 +187,12 @@ namespace l1gt {
z0_t seed_z0;
ap_uint<1> charge;
ap_uint<2> type;
iso_t isolation;
tau_rawid_t isolation;
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just confirming this is a fix, which synchronises the Tau object definition to what's in the interface document?

Copy link

@qvyz qvyz Apr 17, 2023 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @Duchstf ,@EmyrClement, Elias from the GT here, we were just gazing over these "tau_rawid_t". We were wondering if this parameter has any unit, and how it is intended to be "cut" on?

We found this line:

cmssw/L1Trigger/Phase2L1ParticleFlow/plugins/L1NNTauProducer.cc

Line 290 in 7c2e09f

l1ctTau.hwRawId = ap_uint<10>(NN * 1024); //NN Output is ap_fixed<16, 8> so need to cast.

and were wondering if these two are related and if the conversion is correct?
Thanks in advance!
All the best Elias

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would like to add, that the in the menu we would want to select all taus with "passLooseNN()". Can you clarify this point too? Thank you @EmyrClement and @Duchstf

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Duchstf this is my understanding, and you should confirm if it's correct:

  • passLooseNN() is a bool and corresponds to NN score (which is between 0 and 1) is > 0.28
  • l1ct:: hwRawId, which is a ap_uint<10>, corresponds to NN score * 1024 (i.e. from 0->1024)
  • l1gt:: isolation corresponds to l1ct:: hwRawId, and is also a ap_uint<10>

So to apply passLooseNN() on the GT object, I think it's l1gt::isolation > ~286.

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Duchstf From your thumbs up, I think you agree with my suggestion. Can you also confirm that the iso_t -> tau_rawid_t is just a fix to synchronise with the GT format?

ap_uint<2> id0;
ap_uint<2> id1;

static const int BITWIDTH = 128;
inline ap_uint<BITWIDTH> pack() const {
inline ap_uint<BITWIDTH> pack_ap() const {
ap_uint<BITWIDTH> ret;
unsigned int start = 0;
pack_into_bits(ret, start, valid);
Expand All @@ -204,6 +206,44 @@ namespace l1gt {
pack_into_bits(ret, start, id1);
return ret;
}

inline PackedTau pack() const {
PackedTau packed;
ap_uint<BITWIDTH> bits = this->pack_ap();
packed[0] = bits(63, 0);
packed[1] = bits(127, 64);
return packed;
}

inline static Tau unpack_ap(const ap_uint<BITWIDTH> &src) {
Tau ret;
ret.initFromBits(src);
return ret;
}

inline static Tau unpack(const PackedTau &src) {
ap_uint<BITWIDTH> bits;
bits(63, 0) = src[0];
bits(127, 64) = src[1];

return unpack_ap(bits);
}

inline void initFromBits(const ap_uint<BITWIDTH> &src) {
unsigned int start = 0;
unpack_from_bits(src, start, valid);
unpack_from_bits(src, start, v3.pt);
unpack_from_bits(src, start, v3.phi);
unpack_from_bits(src, start, v3.eta);
unpack_from_bits(src, start, seed_pt);
unpack_from_bits(src, start, seed_z0);
unpack_from_bits(src, start, charge);
unpack_from_bits(src, start, type);
unpack_from_bits(src, start, isolation);
unpack_from_bits(src, start, id0);
unpack_from_bits(src, start, id1);
}

}; // struct Tau

struct Electron {
Expand Down
19 changes: 19 additions & 0 deletions DataFormats/L1TParticleFlow/interface/taus.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@

#include "DataFormats/L1TParticleFlow/interface/datatypes.h"
#include "DataFormats/L1TParticleFlow/interface/bit_encoding.h"
#include "DataFormats/L1TParticleFlow/interface/gt_datatypes.h"

namespace l1ct {

Expand Down Expand Up @@ -87,6 +88,7 @@ namespace l1ct {
pack_into_bits(ret, start, hwIsoOrMVA);
return ret;
}

inline static Tau unpack(const ap_uint<BITWIDTH> &src) {
Tau ret;
unsigned int start = 0;
Expand All @@ -102,6 +104,23 @@ namespace l1ct {
unpack_from_bits(src, start, ret.hwIsoOrMVA);
return ret;
}

l1gt::Tau toGT() const {
l1gt::Tau t;
t.valid = hwPt != 0;

t.v3.pt = CTtoGT_pt(hwPt);
t.v3.phi = CTtoGT_phi(hwPhi);
t.v3.eta = CTtoGT_eta(hwEta);

t.seed_pt = hwSeedPt;
t.seed_z0 = hwSeedZ0;
t.charge = hwCharge;

t.type = hwType;
t.isolation = hwRawId;
return t;
}
};

inline void clear(Tau &c) { c.clear(); }
Expand Down
8 changes: 6 additions & 2 deletions DataFormats/L1TParticleFlow/src/PFTau.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,8 @@
#include "DataFormats/L1TParticleFlow/interface/PFTau.h"

l1t::PFTau::PFTau(const PolarLorentzVector& p, float iso, float fulliso, int id, int hwpt, int hweta, int hwphi)
: L1Candidate(p, hwpt, hweta, hwphi, /*hwQuality=*/int(0)), iso_(iso), fullIso_(fulliso), id_(id) {}
l1t::PFTau::PFTau(
const PolarLorentzVector& p, float NNValues[80], float iso, float fulliso, int id, int hwpt, int hweta, int hwphi)
: L1Candidate(p, hwpt, hweta, hwphi, /*hwQuality=*/int(0)), iso_(iso), fullIso_(fulliso), id_(id) {
for (int i0 = 0; i0 < 80; i0++)
NNValues_[i0] = NNValues[i0]; // copy the array of NN inputs
}
3 changes: 2 additions & 1 deletion DataFormats/L1TParticleFlow/src/classes_def.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,8 @@
<class name="edm::RefVector<l1t::PFJetCollection>" />
<class name="std::vector<edm::Ref<std::vector<l1t::PFJet>,l1t::PFJet,edm::refhelper::FindUsingAdvance<std::vector<l1t::PFJet>,l1t::PFJet> > >" />

<class name="l1t::PFTau" ClassVersion="4">
<class name="l1t::PFTau" ClassVersion="5">
<version ClassVersion="5" checksum="549313204"/>
<version ClassVersion="4" checksum="2045755946"/>
<version ClassVersion="3" checksum="1410157160"/>
</class>
Expand Down
1 change: 1 addition & 0 deletions L1Trigger/Phase2L1ParticleFlow/interface/TauNNId.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@ class TauNNId {

void setNNVectorVar();
float EvaluateNN();
float *NNVectorVar() { return NNvectorVar_.data(); }
float compute(const l1t::PFCandidate &iSeed, l1t::PFCandidateCollection &iParts);

private:
Expand Down
1 change: 1 addition & 0 deletions L1Trigger/Phase2L1ParticleFlow/interface/taus/TauNNIdHW.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,6 +147,7 @@ class TauNNIdHW {

void initialize(const std::string &iName, int iNParticles);
void SetNNVectorVar();
input_t *NNVectorVar() { return NNvectorVar_.data(); }
result_t EvaluateNN();
result_t compute(const l1t::PFCandidate &iSeed, std::vector<l1t::PFCandidate> &iParts);
//void print();
Expand Down
15 changes: 7 additions & 8 deletions L1Trigger/Phase2L1ParticleFlow/interface/taus/weights/b1.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,15 +1,14 @@
//Numpy array shape [25]
//Min -1.101188778877
//Max 1.014160394669
//Number of zeros 2
//Min -0.734825849533
//Max 1.288661003113
//Number of zeros 0

#ifndef B1_H_
#define B1_H_

weight_default_t b1[25] = {0.8776568174, -0.0888396949, -0.1198173761, -0.0066847582, -0.0117284302,
-0.0283335019, 0.0000000000, -1.1011887789, -0.0135271018, -0.0323914811,
0.5437909961, -0.0175916012, 0.5357875228, -0.3656347692, 0.2423969060,
1.0141603947, 0.0000000000, -0.7741876245, 0.9614976048, 0.5918464661,
-0.3908625543, -0.2043008506, -0.3004969060, -0.1039064825, 0.5963121057};
weight_default_t b1[25] = {-0.12057505, -0.05409636, 0.27422485, 0.49775919, -0.73482585, 0.44995615, 0.52624124,
-0.71328187, -0.43596983, 0.10772870, -0.68372047, 0.22197038, -0.53673136, -0.00771000,
0.06140821, 1.28866100, -0.12453079, 0.16897179, 0.18858922, -0.17255782, -0.24242370,
-0.21922758, 0.40799412, 0.46138164, 0.85911417};

#endif
24 changes: 12 additions & 12 deletions L1Trigger/Phase2L1ParticleFlow/interface/taus/weights/b2.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,20 +1,20 @@
//Numpy array shape [10]
//Min -0.819882214069
//Max 0.973487198353
//Min -0.380347400904
//Max 0.551839828491
//Number of zeros 0

#ifndef B2_H_
#define B2_H_

weight_default_t b2[10] = {-0.8198822141,
0.7516837120,
0.6504452229,
-0.0292063691,
-0.0308178961,
0.9734871984,
0.1587447226,
-0.3352679014,
-0.0403082110,
0.9563522935};
weight_default_t b2[10] = {0.55183983,
0.36323273,
-0.13108490,
-0.38034740,
0.08559006,
0.01700789,
0.13562575,
-0.32752651,
0.48282012,
-0.15096320};

#endif
24 changes: 12 additions & 12 deletions L1Trigger/Phase2L1ParticleFlow/interface/taus/weights/b3.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,20 +1,20 @@
//Numpy array shape [10]
//Min -0.128828719258
//Max 1.138555169106
//Min -0.936354994774
//Max 0.407682240009
//Number of zeros 0

#ifndef B3_H_
#define B3_H_

weight_default_t b3[10] = {1.0009469986,
-0.0118703200,
0.5378767252,
0.6056469083,
-0.0177963823,
0.3281430006,
-0.0163392760,
1.1385551691,
-0.1288287193,
0.5457931757};
weight_default_t b3[10] = {-0.58549309,
-0.06117089,
-0.24173595,
0.17925857,
-0.93635499,
0.18813914,
0.13134949,
0.04132507,
0.40768224,
0.29987794};

#endif
6 changes: 3 additions & 3 deletions L1Trigger/Phase2L1ParticleFlow/interface/taus/weights/b4.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,11 +1,11 @@
//Numpy array shape [1]
//Min -0.572278082371
//Max -0.572278082371
//Min 0.023343238980
//Max 0.023343238980
//Number of zeros 0

#ifndef B4_H_
#define B4_H_

weight_default_t b4[1] = {-0.5722780824};
weight_default_t b4[1] = {0.02334324};

#endif
Loading