tests/testthat/test_bayesian_decoder.R

In przemyslawj/datatrace: Trace data analysis

library(testthat)
library(dplyr)

context('Tests for the bayesian model')


test_that('to_1dim succeeds', {
  expect_equal(to_1dim(0, 0, nbins.y=20), 1)
  expect_equal(to_1dim(0, 1, nbins.y=20), 2)
  expect_equal(to_1dim(1, 0, nbins.y=20), 21)
}) 

test_that('from1dim succeeds', {
  expect_equal(from_1dim(1, 20), list(x=0, y=0))
  expect_equal(from_1dim(2, 20), list(x=0, y=1))
  expect_equal(from_1dim(21, 20), list(x=1, y=0))
}) 

test_that('timebin.traces success for one trial many cells', {
  df = data.frame(
    animal='A',
    date='2019-01-01',
    exp_title='trial',
    cell_id=c(1,1,1,2,2,2),
    trial_id='1',
    trial=1,
    timestamp=c(1,2,11,1,2,12),
    trace=c(0,2,2,1,3,4),
    deconv_trace=0,
    is.event=FALSE,
    velocity=1,
    dist_reward0=0,
    dist_reward1=0,
    atReward0=0,
    atReward1=0,
    arrivedAtReward=0,
    dist=0,
    smooth_trans_x=1,
    smooth_trans_y=2
  )
  actual.binned = timebin.traces(data.table(df), timebin.dur.msec=10)
  actual.binned = actual.binned %>% arrange(cell_id, time_bin)
  expect_equal(nrow(actual.binned), 4)
  expect_equal(actual.binned$cell_id, c(1,1,2,2))
  expect_equal(actual.binned$time_bin, c(0,1,0,1))
  expect_equal(actual.binned$trace, c(1,2,2,4))
  expect_equal(actual.binned$smooth_trans_x, rep(1,4))
  expect_equal(actual.binned$smooth_trans_y, rep(2,4))
})

test_that('timebin.traces success for two trials', {
  df = data.frame(
    animal='A',
    date='2019-01-01',
    exp_title='trial',
    cell_id=rep(1,6),
    trial_id=rep(c('trial1','trial2'), each=3),
    trial=rep(c(1,2), each=3),
    timestamp=c(1,2,21,1,2,12),
    trace=c(0,2,2,1,3,4),
    deconv_trace=0,
    is.event=FALSE,
    velocity=1,
    dist_reward0=0,
    dist_reward1=0,
    atReward0=0,
    atReward1=0,
    arrivedAtReward=0,
    dist=0,
    smooth_trans_x=1,
    smooth_trans_y=2
  )
  actual.binned = timebin.traces(data.table(df), timebin.dur.msec=10)
  actual.binned = actual.binned %>% arrange(cell_id, time_bin)
  expect_equal(nrow(actual.binned), 4)
  expect_equal(actual.binned$cell_id, rep(1,4))
  expect_equal(actual.binned$time_bin, c(0,2,3,4))
  expect_equal(actual.binned$trace, c(1,2,2,4))
  expect_equal(actual.binned$smooth_trans_x, rep(1,4))
  expect_equal(actual.binned$smooth_trans_y, rep(2,4))
})

test_that("bin.responses success for two cells", {
  df = data.frame(
    animal='A',
    date='2019-01-01',
    cell_id=rep(c(1,2),each=4),
    trial_id='trial1',
    time_bin=rep(1:4),
    trace=c(1:4, 9:6),
    x=1:8,
    y=1
  )
  df$cell_id = as.factor(df$cell_id)
  df = data.table(df)
  quantile.fractions=c(0.5, 1.0)
  actual.binned = bin.responses(df,
                                get.bin.thresholds.fun=get.quantiles.fun(quantile.fractions))
  actual.binned = actual.binned %>% arrange(cell_id, time_bin)
  expect_equal(nrow(actual.binned), 8)
  expect_equal(actual.binned$response_bin, c(1, 1, 2, 2, 2, 2, 1, 1))
})

test_that('create.discrete.bayes success for one cell', {
  df = data.frame(
    animal='A',
    date='2019-01-01',
    cell_id=1,
    time_bin=1:6,
    response_bin=c(2,2,1,1,1,2),
    bin.xy=      c(1,1,2,2,4,4)
  )
  
  model = create.discrete.bayes(df, nstim.bins=4)
  
  likelihood = model$likelihood
  expect_equal(dim(likelihood), c(4,1,2))
  expect_equal(likelihood[,1,1], c(0.25, 0.75, NA, 0.5))
  expect_equal(likelihood[,1,2], c(0.75, 0.25, NA, 0.5))
  
  expected.prior = rep(1/3,4)
  expected.prior[3] = 0.0
  expect_equal(model$prior, expected.prior)
})


test_that('bayesmax success with equal prior and same cells', {
  likelihoodM = array(0.1, dim=c(4,2,2))
  colnames(likelihoodM) = c('1', '2')
  likelihoodM[3,'1',1] = 0.9
  likelihoodM[3,'2',2] = 0.9
  
  model.bayes = list(prior=rep(0.25, 4),
                     likelihood=likelihoodM)
  pv = c(`1`=1, `2`=2)
  #res = bayes.max.s2(model.bayes, pv)
  res = bayesmax(model.bayes$prior, model.bayes$likelihood, pv)
  expect_equal(3, res$s)
})

test_that('bayesmax success with equal prior and different cells', {
  likelihoodM = array(0.1, dim=c(4,2,2))
  colnames(likelihoodM) = c('1', '2')
  likelihoodM[1,'1',1] = 0.9
  likelihoodM[4,'2',2] = 0.7
  likelihoodM[4,'2',1] = 0.3
  
  model.bayes = list(prior=rep(0.25, 4),
                     likelihood=likelihoodM)
  pv = c(`2`=1, `3`=2)
  #res = bayes.max.s2(model.bayes, pv)
  res = bayesmax(model.bayes$prior, model.bayes$likelihood, pv)
  expect_equal(4, res$s)
})

test_that('bayesmax success with unequal prior', {
  likelihoodM = array(0.25, dim=c(4,2,2))
  colnames(likelihoodM) = c('1', '2')
  likelihoodM[3,'1',1] = 0.7
  likelihoodM[3,'1',2] = 0.3
  
  model.bayes = list(prior=c(0.7, rep(0.1, 3)),
                     likelihood=likelihoodM)
  pv = c(`1`=1)
  res = bayesmax(model.bayes$prior, model.bayes$likelihood, pv)
  #res = bayes.max.s2(model.bayes, pv)
  expect_equal(1, res$s)
})