szilard

## Stochastic Variational Deep Kernel Learning
## paper: https://arxiv.org/abs/1611.00336
## code+data from the authors (thanks!!!): https://people.orie.cornell.edu/andrew/code/#SVDKL


## get data + prepare sample authors used for evaluation

wget https://people.orie.cornell.edu/andrew/code/svdklcode.zip
unzip  svdklcode.zip

szilard

##install.packages("cranlogs")

library(data.table)
library(cranlogs)


##caret/models/file
## grep "library =" * | sed 's/.*=//' | sed 's/c(//' | sed 's/),/,/' | grep -v NULL | sed 's/,.*$/,/' | sort | uniq | tr -d '\n'

caret_pkgs <- c("rpart", "C50", "CHAID", "Cubist", "FCNN4R", "HDclassif", "HiDimDA", "KRLS", "LiblineaR", 

szilard

library(data.table)
library(rpart)

d_train <- fread("https://s3.amazonaws.com/benchm-ml--main/train-0.1m.csv") 

md <- rpart(ifelse(dep_delayed_15min=="Y",1,0) ~ ., d_train, 
            control = rpart.control(cp = 0.001))

plotcp(md)
printcp(md)

szilard

data.table vs SparkR

group-by aggregate on 100M records (1M groups)


data.table 6.5 sec (without key) / 1.3 sec (with key) - all 1 core

SparkR cached 200 sec (8 cores)
30x / 150x   ( 240x / 1200x per core)

szilard

## training a model

library(h2o)
h2o.init(nthreads = -1)

dx_train <- h2o.importFile("https://s3.amazonaws.com/benchm-ml--main/train-0.1m.csv")

md_rf <- h2o.randomForest(x = 1:(ncol(dx_train)-1), y = ncol(dx_train), training_frame = dx_train, 
                    model_id = "h2o_RF",
                    ntrees = 100, max_depth = 10, nbins = 100)

szilard

library(data.table)
library(ROCR)
library(lightgbm)

set.seed(123)

d_train <- fread("/var/data/bm-ml/train-0.1m.csv")
d_test <- fread("/var/data/bm-ml/test.csv")

szilard

## partial credit :) to @earino for the idea

library(lightgbm)
library(data.table)
library(ROCR)

d0_train <- fread("/var/data/bm-ml/train-10m.csv") 
d0_test <- fread("/var/data/bm-ml/test.csv") 
d0 <- rbind(d0_train, d0_test)

szilard

# OpenML Benchmarking Suites and the OpenML100
# https://arxiv.org/abs/1708.03731
# https://www.openml.org/s/14/data

library(OpenML)
ids <- getOMLStudy('OpenML100')$data$data.id
dsall <- listOMLDataSets()
sum(dsall$data.id %in% ids)   ## 96???
ds <- dsall[dsall$data.id %in% ids,]

szilard

## https://github.com/EpistasisLab/penn-ml-benchmarks

## pip install pmlb

import numpy as np
from pmlb import fetch_data
from pmlb import dataset_names

x = np.zeros(len(dataset_names))
for i, dn in enumerate(dataset_names):

szilard

#include <stdio.h>
#include <stdlib.h>

#define N 128
#define B0 100
#define R 1000000
#define M 1000

int cmpfunc (const void * a, const void * b)
{