| ######################################## | |
| ## # Train BDT and output model .json ## | |
| ######################################## | |
| import numpy as np | |
| from sklearn.model_selection import train_test_split | |
| import sys | |
| import os | |
| import xgboost as xgb | |
| ### Define test and train matrices somehow |
| import os | |
| import sys | |
| import commands | |
| import datetime | |
| import time | |
| """ | |
| 1. Visit this exact URL in Google Chrome | |
| https://cmswbm.cern.ch/cmsdb/servlet/ConditionBrowser?DISPLAY=1&SUBMIT=1D&OPTION=vs_time&STEP=3&BEGIN=2017.10.29_09:50:16&END=2017.10.31_09:50:16&PRESCALER=1&cms_omds_lb.CMS_CSC_PVSS_COND.CSC_HV.CURRENT..M13_C03..10=1&cms_omds_lb.CMS_CSC_PVSS_COND.CSC_HV.CURRENT..M13_C03..11=1& |
| let g:pathogen_disabled = [] " disable certain pathogen plugins ... useful to debug | |
| " load plugins before screwing with settings | |
| " https://github.com/tpope/vim-commentary | |
| " https://github.com/tpope/vim-surround | |
| " https://github.com/tpope/vim-fugitive | |
| " https://github.com/scrooloose/nerdtree | |
| " https://github.com/haya14busa/incsearch.vim | |
| " https://github.com/haya14busa/incsearch-fuzzy.vim # z/ to fuzzy search |
| import os | |
| import pickle | |
| import pydot | |
| import keras | |
| from keras.models import load_model | |
| """ | |
| brew install graphviz | |
| pip install pydot |
Execute the commands chunk by chunk because I haven't tested this as a whole
# setup environment
cd /cvmfs/cms.cern.ch/slc7_amd64_gcc630/cms/cmssw/CMSSW_10_2_0_pre6/ ; cmsenv ; cd -
# make containment folder somewhere and get vi
mkdir vim8
cd vim8
git clone https://github.com/vim/vim
| /* | |
| Copy this content as a bookmarklet. | |
| Visit https://cmswbmoffshift.web.cern.ch/cmswbmoffshift/api/query and make sure you're authenticated (should say "Query ID missing?") | |
| Execute the bookmark. | |
| ??? | |
| Profit. | |
| Note: will need to search for and update the following strings below (whenever conditions change): "PromptReco-v3", "Run2018", "319750" | |
| The queries will return at most 1k rows (usually 4k runs or so) so make sure the run number threshold isn't too small! | |
| Well, this won't be useful after 2018... |
| import sys | |
| import time | |
| import numba | |
| import numpy as np | |
| import awkward | |
| from awkward import JaggedArray | |
| @numba.jit(nopython=True,cache=True) | |
| def numba_min(content,offsets): | |
| result = np.zeros(len(offsets)-1,dtype=content.dtype) |
| from __future__ import print_function | |
| """ | |
| stolen from https://github.com/dmwm/CRABClient/blob/master/src/python/CRABClient/JobType/UserTarball.py | |
| """ | |
| import os | |
| import glob | |
| import tarfile | |
| import tempfile |
| # in some versions of xgboost, can't easily dump to json, so dump to the "plaintext" | |
| # format, and then do some string parsing to make a dictionary and then json | |
| # next, convert the json to a c++ function (nested ternary statements for the tree branching) | |
| # and don't forget the sigmoid at the end to get probabilities! | |
| # let's say you have | |
| # ... | |
| # bst = xgb.train( param.items(), dtrain, num_round, evallist, early_stopping_rounds=20 ) | |
| # feature_names = ["njets", "nbtags", "met"] |