thrash.R

#### Thrash for ICA extended abstract and for processing network data



# 1D ----------------------------------------------------------------------

entropy1D = function(v1, bin.width = 0.1) {
  # preparation of empty vector of distances
  l = length(v1)  # Length of file/number of agents in simulation
  ld = (l * (l - 1)) / 2  # Length of 'd' is number of pair of agents, according equation: N * (N - 1) / 2
  d = rep(NA, ld)  # We construct empty vector of length 'ld'
  c = 0  # Counter which will be updated inside of the following cycles so to be able points to the first empty cell in 'd'.
  # Note: we start 'c' with 0 because we will start with updating 'c'.

  # We need two for cycles for going over all pairs of agents and measures their distance in 2D space
  for (i in 1:(l - 1)) {  # We start first cycle with the first agent and end with the next to the last.
    for (j in (i + 1):l) {
      c = c + 1  # Updating counter
      d[c] = abs(v1[i] - v1[j])  # Euclidean distance
    }
  }

  # Now we use histogram function to separate vector into bins and
  # transform it into counts of these bins.
  cnts = hist(d, breaks = seq(0, 2.11, bin.width))$counts  # Transformation
  cnts = cnts[cnts>0]  # Filtering out 0s.

  # Now we transform counts into relative freqencies:
  frqs = cnts / ld

  # Now we transform them into product of relative frequency and its logarithm:
  ents = frqs * log(frqs, 2)

  # Now we need the value of maximum entropy with given granularity:
  bin.num = ceiling(2 / bin.width)
  max.ent = log(1 / bin.num, 2)

  # Here in clean environment we prepare all coefficients:
  p_ent =  round(100 - (sum(ents) / (max.ent) * 100), 1)
  p_oneGroup = round(sqrt(cnts[1] * 2) / l * 100, 1)
  n_groups = round(l * l / (l + (2 * cnts[1])), 2)
  p_lengths =  round(frqs[1] * 100, 1)
  SD = round(sd(d), 3)
  DI = round(mean(d) / 2 * 100, 3)

  # Returning back four parameters as one string:
  #print(paste(p_ent, p_oneGroup, n_groups, p_lengths, SD, sep = "_"))
  paste(p_ent, p_oneGroup, n_groups, p_lengths, SD, DI, sep = "_")
}

# dfp = dfb %>% filter(Dimensions == 1, Sim_ID <= 4) %>% group_by(Sim_ID)
# de = entropy1D(dfp$Opinion1_Final, 0.1)
# de



a = Sys.time()
dfy = dfb %>%
  # filter(Sim_ID <= 10) %>%
  filter(Dimensions == 1) %>%
  group_by(Sim_ID, Seed, Population, Random_links, Close_links, Dimensions, Boundary, Boundary_method, P_speaking,  Mode) %>%
  summarise(ent = entropy1D(Opinion1_Final, 0.1)) %>%
  separate(ent,
           into = c("Far_from_entropy", "One_group_size",
                    "Number_of_equal_groups", "Zero_lenghts", "SD", "DI"),
           sep = "_", convert = T)
b = Sys.time()
b - a

# Saving processed meta indicators:
write_csv(dfy, "Sims02_processed_1D.csv")



# 2D ----------------------------------------------------------------------

entropy2D = function(v1, v2, bin.width = 0.1) {
  # preparation of empty vector of distances
  l = length(v1)  # Length of file/number of agents in simulation
  ld = (l * (l - 1)) / 2  # Length of 'd' is number of pair of agents, according equation: N * (N - 1) / 2
  d = rep(NA, ld)  # We construct empty vector of length 'ld'
  c = 0  # Counter which will be updated inside of the following cycles so to be able points to the first empty cell in 'd'.
  # Note: we start 'c' with 0 because we will start with updating 'c'.

  # We need two for cycles for going over all pairs of agents and measures their distance in 2D space
  for (i in 1:(l - 1)) {  # We start first cycle with the first agent and end with the next to the last.
    for (j in (i + 1):l) {
      c = c + 1  # Updating counter
      d1 = v1[i] - v1[j]  # distance in D1
      d2 = v2[i] - v2[j]  # distance in D2
      d[c] = sqrt((d1 ^ 2) + (d2 ^ 2))  # Euclidean distance
    }
  }

  # Now we use histogram function to separate vector into bins and
  # transform it into counts of these bins.
  cnts = hist(d, breaks = seq(0, 3, bin.width))$counts  # Transformation
  cnts = cnts[cnts>0]  # Filtering out 0s.

  # Now we transform counts into relative freqencies:
  frqs = cnts / ld

  # Now we transform them into product of relative frequency and its logarithm:
  ents = frqs * log(frqs, 2)

  # Now we need the value of maximum entropy with given granularity:
  bin.num = ceiling(2.82 / bin.width)
  max.ent = log(1 / bin.num, 2)

  # Here in clean environment we prepare all coefficients:
  p_ent =  round(100 - (sum(ents) / (max.ent) * 100), 1)
  p_oneGroup = round(sqrt(cnts[1] * 2) / l * 100, 1)
  n_groups = round(l * l / (l + (2 * cnts[1])), 2)
  p_lengths =  round(frqs[1] * 100, 1)
  SD = round(sd(d), 3)
  DI = round(mean(d) / sqrt(8) * 100, 3)

  # Returning back four parameters as one string:
  #print(paste(p_ent, p_oneGroup, n_groups, p_lengths, SD, sep = "_"))
  paste(p_ent, p_oneGroup, n_groups, p_lengths, SD, DI, sep = "_")
}

a = Sys.time()
dfy = dfb %>%
  # filter(Sim_ID <= 72) %>%
  filter(Dimensions == 2) %>%
  group_by(Sim_ID, Seed, Population, Random_links, Close_links, Dimensions, Boundary, Boundary_method, P_speaking,  Mode) %>%
  summarise(ent = entropy2D(Opinion1_Final, Opinion2_Final, 0.1)) %>%
  separate(ent,
           into = c("Far_from_entropy", "One_group_size",
                    "Number_of_equal_groups", "Zero_lenghts", "SD", "DI"),
           sep = "_", convert = T)
b = Sys.time()
b - a

# Saving processed meta indicators:
write_csv(dfy, "Sims02_processed_2D.csv")




# ======================= OLD CODE ====================================

# Since data-files are still huge, we have to combine separately data with POPULATIONS 513, 257, 129,
# so we firstly need cycle iterating over ´Population´:
for (k in c(129, 257, 513)) {

  # Since data are long we have to do them seed by seed.
  # So we need secondly the cycle iterating over seeds:
  for (j in 1:10) {

    # We take a vector with filenames of all simulation results from directory 'Sims':
    d = dir(path = "Sims", pattern = paste0("Sims02_", j,"_", k, "_*"))

    # We read the first file and transform it accordingly for the storing in master datafile:
    df = read_csv(paste0("Sims/", d[37])) %>%  # NOTE: 37 is important -- we have to find the first simulation with 2 dimensions of opinion space to more smoothly add cases of data-files from other simulations
      select(-starts_with("Nei")) %>%  # NOTE: without comp. cluster or better code or data storing politics we can't read in also the network data, files are then too huge and personal computer is not able to red, add and store updated dataframes, so for now, we have to information on networks cut out. Now the option is to agree on what we need to prapare from the network data, prepare it and store by each agent or by each simulation.
      mutate(Meta = d[37]) %>%  # We store file name as variable, since the name contains all meta data
      separate(
        Meta, sep = "_",
        into = c(  # We separate meta info into distinctive variables:
          "Sim", "Seed","Population", "Random_links","Close_links", "Dimensions",
          "Updating", "Boundary","Boundary_method", "P_speaking", "Mode")) %>%
      mutate(Mode = str_sub(Mode, end = -5),  # We cut out last 4 characters from new variable mode, since they are ".csv"
             Phase = if_else(Step==0, "Start", "Final"),  # We prepare variable coining final and starting state of agent/simulation
             Step = max(Step)) %>%  # For the next step of data-management we need var ´Step´ constant
      pivot_wider(id_cols = c(ID:Uncertainty, Step:Phase),  # Since a lot of data on final state are redundant -- they are constant for whole simulation -- we create opinion variables for starting and final state and the rest of data is same, so we save a lot of space!
                  names_from = Phase, names_glue = "{.value}_{Phase}",  # We identify cases by all data, except Opinion vars.
                  values_from = starts_with("Opinion")) %>%
      relocate(Sim:Mode, .after = ID) %>% relocate(Step, .after = Mode) %>%
      relocate(starts_with("Opinion"), .after = Uncertainty)  # We logically relocate vars.

    # We save processed data-file.
    # NOTE: since this is the first record, by 'append = F' we erase potential older versin of .csv file.
    write_csv(df, paste0("Sims02_", j,"_", k, ".csv"), append = F)

    # We iterate over whole vector -- we prepare next file for inclusion and
    # then append it into .csv file.
    # NOTE: Code is the very same as the code for the first record,
    #       annotated are only differing parts of code.
    for (i in c(1:36, 38:length(d))) {
      # for (i in c(34:36, 38:39)) {   #
      df2 = read_csv(paste0("Sims/", d[i])) %>%
        select(-starts_with("Nei")) %>%  # NOTE: After commenting this line out the data on network structure stay in the file.
        mutate(Meta = d[i]) %>%
        separate(
          Meta, sep = "_",
          into = c(
            "Sim", "Seed","Population", "Random_links","Close_links", "Dimensions",
            "Updating", "Boundary","Boundary_method", "P_speaking", "Mode")) %>%
        mutate(Mode = str_sub(Mode, end = -5),
               Phase = if_else(Step==0, "Start", "Final"),
               Step = max(Step)) %>%
        pivot_wider(id_cols = c(ID:Uncertainty, Step:Phase),
                    names_from = Phase, names_glue = "{.value}_{Phase}",
                    values_from = starts_with("Opinion")) %>%
        relocate(Sim:Mode, .after = ID) %>% relocate(Step, .after = Mode) %>%
        relocate(starts_with("Opinion"), .after = Uncertainty)

      # Storing record into main .csv file:
      # NOTE: Note 'append = T', it means that we append next record to the existing .csv file.
      write_csv(df2, paste0("Sims02_", j,"_", k, ".csv"), append = T)
    }
  }
}



# Reading files back and joining them: ------------------------------------

# We do it for sure step by step.
# Firstly we make a file for each population size and them join them into one master file.

# Joining files for population size 129:
df129 = read_csv("Sims/Sims02_1_129.csv") %>%
  add_row(read_csv("Sims/Sims02_2_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_3_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_4_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_5_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_6_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_7_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_8_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_9_129.csv")) %>%
  add_row(read_csv("Sims/Sims02_10_129.csv"))
write_csv(df129, "Sims02_pop129.csv")


# Joining files for population size 257:
df257 = read_csv("Sims/Sims02_1_257.csv") %>%
  add_row(read_csv("Sims/Sims02_2_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_3_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_4_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_5_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_6_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_7_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_8_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_9_257.csv")) %>%
  add_row(read_csv("Sims/Sims02_10_257.csv"))
write_csv(df257, "Sims02_pop257.csv")


# Joining files for population size 513:
df513 = read_csv("Sims/Sims02_1_513.csv") %>%
  add_row(read_csv("Sims/Sims02_2_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_3_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_4_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_5_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_6_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_7_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_8_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_9_513.csv")) %>%
  add_row(read_csv("Sims/Sims02_10_513.csv") %>% filter(Dimensions < 3) %>% mutate(Opinion2_Final = as.numeric(Opinion2_Final)))
write_csv(df513, "Sims02_pop513.csv")


# Joining population files into the master file:
dfa = read_csv("Sims02_pop129.csv") %>% select(-starts_with("Nei")) %>%
  add_row(read_csv("Sims02_pop257.csv")) %>%
  add_row(read_csv("Sims02_pop513.csv"))
write_csv(dfa, "Sims02_all.csv")


# Reding the master file back:
dfa = read_csv("Sims02_all.csv")


# Preparing data for the analysis:
dfx = dfa %>% select(-Sim) %>%
  group_by(Seed, Population, Random_links, Close_links, Dimensions, Boundary, Boundary_method, P_speaking,  Mode) %>%
  mutate(Sim_ID = cur_group_id()) %>% relocate(Sim_ID, .before = ID)
write_csv(dfx, "Sims02_all_new.csv")

# Reading back:
dfb = read_csv("Sims02_all_new.csv")
dfx = dfb %>% filter(Dimensions == 2) %>% group_by(Sim_ID) %>%
  summarise(op1 = sd(Opinion1_Final), op2 = sd(Opinion2_Final))
ggplot(dfx, aes(x = op1, y = op2)) + geom_point(alpha = 0.1) + theme_minimal()
ggplot(dfx, aes(x = op1, y = op2)) + geom_density_2d() + theme_minimal()
ggplot(dfx, aes(x = op1)) + geom_density() + theme_minimal()
ggplot(dfx, aes(x = op2)) + geom_density() + theme_minimal()

dfy = dfb %>% filter(Sim_ID == 39)

# dfp = dfb %>% filter(Dimensions == 2, Sim_ID < 40) %>% group_by(Sim_ID)
# de = entropy2D(dfp$Opinion1_Final, dfp$Opinion2_Final, 0.1)
# de


# Reading in and storing data from HK-benchmark -----------------------------------------

# As with main data, we firstly need cycle iterating over ´Population´:
for (k in c(513)) { # 129, 257,

  # As with main data, we need secondly the cycle iterating over seeds:
  for (j in 1:73) {

    # We take a vector with filenames of all simulation results from directory 'Sims':
    d = dir(path = "Sims", pattern = paste0("Sims02_", j,"_", k, "_*"))

    # We read the first file and transform it accordingly for the storing in master datafile:
    df = read_csv(paste0("Sims/", d[1])) %>%
      select(-starts_with("Nei")) %>%  # NOTE: without comp. cluster or better code or data storing politics we can't read in also the network data, files are then too huge and personal computer is not able to red, add and store updated dataframes, so for now, we have to information on networks cut out. Now the option is to agree on what we need to prapare from the network data, prepare it and store by each agent or by each simulation.
      mutate(Meta = d[1]) %>%  # We store file name as variable, since the name contains all meta data
      separate(
        Meta, sep = "_",
        into = c(  # We separate meta info into distinctive variables:
          "Sim", "Seed","Population", "Random_links","Close_links", "Dimensions",
          "Updating", "Boundary","Boundary_method", "P_speaking", "Mode")) %>%
      mutate(Mode = str_sub(Mode, end = -5),  # We cut out last 4 characters from new variable mode, since they are ".csv"
             Phase = if_else(Step==0, "Start", "Final"),  # We prepare variable coining final and starting state of agent/simulation
             Step = max(Step)) %>%  # For the next step of data-management we need var ´Step´ constant
      pivot_wider(id_cols = c(ID:Uncertainty, Step:Phase),  # Since a lot of data on final state are redundant -- they are constant for whole simulation -- we create opinion variables for starting and final state and the rest of data is same, so we save a lot of space!
                  names_from = Phase, names_glue = "{.value}_{Phase}",  # We identify cases by all data, except Opinion vars.
                  values_from = starts_with("Opinion")) %>%
      relocate(Sim:Mode, .after = ID) %>% relocate(Step, .after = Mode) %>%
      relocate(starts_with("Opinion"), .after = Uncertainty)  # We logically relocate vars.

    # We save processed data-file.
    # NOTE: since this is the first record, by 'append = F' we erase potential older versin of .csv file.
    write_csv(df, paste0("HK02_", j,"_", k, ".csv"), append = F)

    # We iterate over whole vector -- we prepare next file for inclusion and
    # then append it into .csv file.
    # NOTE: Code is the very same as the code for the first record,
    #       annotated are only differing parts of code.
    for (i in c(2:length(d))) {
      df2 = read_csv(paste0("Sims/", d[i])) %>%
        select(-starts_with("Nei")) %>%  # NOTE: After commenting this line out the data on network structure stay in the file.
        mutate(Meta = d[i]) %>%
        separate(
          Meta, sep = "_",
          into = c(
            "Sim", "Seed","Population", "Random_links","Close_links", "Dimensions",
            "Updating", "Boundary","Boundary_method", "P_speaking", "Mode")) %>%
        mutate(Mode = str_sub(Mode, end = -5),
               Phase = if_else(Step==0, "Start", "Final"),
               Step = max(Step)) %>%
        pivot_wider(id_cols = c(ID:Uncertainty, Step:Phase),
                    names_from = Phase, names_glue = "{.value}_{Phase}",
                    values_from = starts_with("Opinion")) %>%
        relocate(Sim:Mode, .after = ID) %>% relocate(Step, .after = Mode) %>%
        relocate(starts_with("Opinion"), .after = Uncertainty)

      # Storing record into main .csv file:
      # NOTE: Note 'append = T', it means that we append next record to the existing .csv file.
      write_csv(df2, paste0("HK02_", j,"_", k, ".csv"), append = T)
    }
  }
}



# Reading files from HK-benchmark and joining them: ------------------------------------

# We do it for sure step by step.
# Firstly we make a file for each population size and them join them into one master file.

# Joining files for population size 129:
# dfa = read_csv("Sims/HK02_1_129.csv") %>%
#   add_row(read_csv("Sims/HK02_2_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_3_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_4_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_5_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_6_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_7_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_8_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_9_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_10_129.csv")) %>%
#   add_row(read_csv("Sims/HK02_1_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_2_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_3_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_4_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_5_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_6_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_7_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_8_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_9_257.csv")) %>%
#   add_row(read_csv("Sims/HK02_10_257.csv"))%>%
#   add_row(read_csv("Sims/HK02_1_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_2_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_3_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_4_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_5_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_6_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_7_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_8_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_9_513.csv")) %>%
#   add_row(read_csv("Sims/HK02_10_513.csv"))

# Firstly we need to create empty tibble:
dfc = tibble(ID = NA, Sim = NA, Seed = NA, Population = NA, Random_links = NA, Close_links = NA,
             Dimensions = NA, Updating = NA, Boundary = NA, Boundary_method = NA,
             P_speaking = NA, Mode = NA, Step = NA, Uncertainty = NA,
             Opinion1_Start = NA, Opinion1_Final = NA)
# NOTE: Later we have to erase the first row full of NAs

# As with main data, we firstly need cycle iterating over ´Population´:
for (k in c(129, 257, 513)) {

  # As with main data, we need secondly the cycle iterating over seeds:
  for (j in 1:73) {

    # We construct the name of the processed file and then add it:
    dfc = dfc %>% add_row(read_csv(paste0("Sims/HK02_", j, "_", k, ".csv")))
  }
}

# Erasing empty line out:
dfc = slice(dfc, -1)

# Writing pre-processed data:
write_csv(dfc, "HK02_all.csv")


# Reding the master file back:
# dfa = read_csv("HK02_all.csv")


# Preparing data for the analysis:
dfx = dfc %>% select(-Sim) %>%
  group_by(Seed, Population, Random_links, Close_links, Dimensions, Boundary, Boundary_method, P_speaking,  Mode) %>%
  mutate(Sim_ID = 13000 + cur_group_id()) %>% relocate(Sim_ID, .before = ID)
write_csv(dfx, "HK02_all_new.csv")

dfa = read_csv("HK02_all_new.csv")


# Processing HK-benchmark -------------------------------------------------

dfp = dfa %>% filter(Dimensions == 1, Sim_ID <= 13004) %>% group_by(Sim_ID)
de = entropy1D(dfp$Opinion1_Final, 0.1)
de



a = Sys.time()
dfz = dfa %>%
  # filter(Sim_ID <= 10) %>%
  filter(Dimensions == 1) %>%
  group_by(Sim_ID, Seed, Population, Random_links, Close_links, Dimensions, Boundary, Boundary_method, P_speaking,  Mode) %>%
  summarise(ent = entropy1D(Opinion1_Final, 0.1)) %>%
  separate(ent,
           into = c("Far_from_entropy", "One_group_size",
                    "Number_of_equal_groups", "Zero_lenghts", "SD", "DI"),
           sep = "_", convert = T)
b = Sys.time()
b - a

# Saving processed meta indicators:
write_csv(dfz, "Sims02_processed_HK.csv")





# Joining files -----------------------------------------------------------

# Firstly, we need to join files/add rows from different entropy dimensions,
# later we add/join also files fro HK-benchmarking experiment
df = read_csv("Sims02_processed_1D.csv") %>%
  add_row(read_csv("Sims02_processed_2D.csv")) %>%
  add_row(read_csv("Sims02_processed_HK.csv")) %>%
  filter(P_speaking > 0.4) %>%
  mutate(Nei_size = round((Close_links * 2) / (Population - 1) * 100, 1)) %>%
  relocate(Nei_size, .after = Population)

# Saving all processed meta indicators:
write_csv(df, "Sims02_processed_all.csv")
df = read_csv("Sims02_processed_all.csv")
save(df, file = "dataWithoutNetworks_all.RData")

# Entropy graphs ----------------------------------------------------------

ggplot(df, aes(x = Far_from_entropy, y = Zero_lenghts)) +
  facet_grid(cols = vars(Dimensions), rows = vars(Boundary_method)) +
  geom_point(alpha = 0.1) +
  theme_minimal()

ggplot(df, aes(x = Far_from_entropy, y = One_group_size)) +
  facet_grid(cols = vars(Dimensions), rows = vars(Nei_size)) +
  geom_point(alpha = 0.1) +
  scale_x_log10() +
  theme_minimal()

ggplot(df, aes(x = Far_from_entropy, y = Number_of_equal_groups)) +
  facet_grid(cols = vars(Dimensions), rows = vars(Boundary)) +
  geom_point(alpha = 0.1) +
  scale_x_log10() +
  scale_y_log10(breaks = c(seq(1, 7, 2), 10, 20, 30)) +
  theme_minimal()

ggplot(df, aes(x = Far_from_entropy, y = SD)) +
  facet_grid(cols = vars(Dimensions), rows = vars(Population)) +
  geom_point(alpha = 0.1) +
  # scale_x_log10() +
  # scale_y_log10() +
  theme_minimal()


# Graph sketches ----------------------------------------------------------

ggplot(filter(dfa, Boundary_method != "constant" ),
       aes(x = Uncertainty, y = abs(Opinion1_Final - Opinion1_Start))) +
  facet_wrap(vars(Mode)) +
  geom_point(alpha = 0.01) +
  theme_minimal()
ggsave("try0.jpg")


dfa %>% filter(Boundary_method != "constant" & Dimensions == 2 & Population == 129) %>%
  ggplot(aes(x = abs(Opinion1_Final - Opinion1_Start),
             y = abs(Opinion2_Final - Opinion2_Start))) +
  facet_grid(cols = vars(Close_links), rows = vars(Boundary)) +
  geom_point(alpha = 0.01) +
  scale_x_log10() +
  scale_y_log10() +
  theme_minimal()
ggsave("try1.jpg", width = 8, height = 8)


dfa %>% filter(Boundary_method != "constant" & Dimensions == 2) %>%
  mutate(Close_links = round(Close_links * 2) / (Population - 1) * 100, 1) %>%
  ggplot(aes(x = abs(Opinion1_Final - Opinion1_Start),
             y = abs(Opinion2_Final - Opinion2_Start))) +
  facet_grid(cols = vars(Close_links), rows = vars(Boundary)) +
  geom_point(alpha = 0.01) +
  theme_minimal()
ggsave("try2.jpg", width = 14, height = 6)


dfa %>% filter(Boundary_method != "constant" & Dimensions == 2) %>%
  ggplot(aes(x = abs(Opinion1_Final - Opinion1_Start),
             y = abs(Opinion2_Final - Opinion2_Start),
             col = Mode)) +
  facet_grid(cols = vars(Close_links, Population), rows = vars(Boundary, Mode)) +
  geom_point(alpha = 0.01) +
  guides(col = "none") +
  theme_minimal()
ggsave("try3.jpg", width = 18, height = 12)



# ZCU graphs --------------------------------------------------------------

dfz = df %>%
  # filter(Seed < 11, Mode == "openly-listen") %>%
  mutate(NeiX = case_when(
    Nei_size < 10 ~ "<10",
    Nei_size > 10 & Nei_size < 100 ~ "12-50",
    Nei_size == 100 ~ "100"
  ) %>% factor(levels = c("<10", "12-50", "100")),
  P_speaking = factor(P_speaking),
  Nei_size = factor(Nei_size),
  Polar = Far_from_entropy * DI / 50)

ggplot(dfz, aes(x = Far_from_entropy, y = DI, col = P_speaking)) +
  facet_grid(cols = vars(Boundary_method, NeiX), rows = vars(Dimensions, P_speaking)) +
  geom_point(alpha = 0.1, size = 3) +
  # scale_x_log10() +
  # scale_y_log10() +
  guides(color = "none") +
  theme_minimal()
ggsave("zcu.jpg", width = 8, height = 8)

ggplot(dfz, aes(x = Nei_size, y = Polar, col = P_speaking)) +
  facet_grid(cols = vars(Boundary_method, Boundary, Mode), rows = vars(Dimensions, P_speaking, Random_links)) +
  geom_boxplot(alpha = 0.1) +
  geom_jitter(alpha = 0.1, size = 3, width = 0.2) +
  # scale_x_log10() +
  # scale_y_log10() +
  guides(color = "none") +
  theme_minimal()
ggsave("zcu2.jpg", width = 16, height = 16)






# ### This is the end my only friend, the end! ### ------------------------



# Old code ----------------------------------------------------------------

entropy2D_show = function(v1, v2, bin.width = 0.1) {
  # preparation of empty vector of distances
  l = length(v1)  # Length of file/number of agents in simulation
  ld = (l * (l - 1)) / 2  # Length of 'd' is number of pair of agents, according equation: N * (N - 1) / 2
  d = rep(NA, ld)  # We construct empty vector of length 'ld'
  c = 0  # Counter which will be updated inside of the following cycles so to be able points to the first empty cell in 'd'.
  # Note: we start 'c' with 0 because we will start with updating 'c'.

  # We need two for cycles for going over all pairs of agents and measures their distance in 2D space
  for (i in 1:(l - 1)) {  # We start first cycle with the first agent and end with the next to the last.
    for (j in (i + 1):l) {
      c = c + 1  # Updating counter
      d1 = v1[i] - v1[j]  # distance in D1
      d2 = v2[i] - v2[j]  # distance in D2
      d[c] = sqrt((d1 ^ 2) + (d2 ^ 2))  # Euclidean distance
    }
  }

  # Now we use histogram function to separate vector into bins and
  # transform it into counts of these bins.
  cnts = hist(d, breaks = seq(0, 3, bin.width))$counts  # Transformation
  cnts = cnts[cnts>0]  # Filtering out 0s.

  # Now we transform counts into relative freqencies:
  frqs = cnts / ld

  # Now we transform them into product of relative frequency and its logarithm:
  ents = frqs * log(frqs, 2)

  # Now we need the value of maximum entropy with given granularity:
  bin.num = ceiling(2.82 / bin.width)
  max.ent = log(1 / bin.num, 2)
  #max.ent = log(1 / length(ents))


  #print(cnts)
  #print(frqs)
  print(ggplot(tibble(v1, v2), aes(x = v1, y = v2)) +
          geom_jitter(alpha = 0.2, height = bin.width / 10, width = bin.width / 10) +
          scale_x_continuous(breaks = seq(-1, 1, 0.2)) +
          scale_y_continuous(breaks = seq(-1, 1, 0.2)) +
          theme_minimal())
  #print(bin.num)
  #print(max.ent)
  # print(paste("Effective bins:", length(cnts), "out of", bin.num))
  print(paste("Average percentage of lengths close to 0 from theoretical maximum:", round(frqs[1] * 100, 1)))
  print(paste("Percentage of agents enough to cover in one group all lenghts close to 0:", round(sqrt(cnts[1] * 2) / l * 100, 1)))
  print(paste("Minimum number of equal groups capable to cover all lenghts close to 0:", round(l * l / (l + (2 * cnts[1])), 2)))
  print(paste("Number of lengths:", ld, "; Number of agents:", l, "; Number of lenghts close to 0:", cnts[1]))
  print(paste("Percentage of uniform public sphere:", round(100 - (sum(ents) / (max.ent) * 100), 1)))
  # print(max.ent2)
  # print(sum(ents) / (max.ent2) * 100)
  # print(tibble(cnts, frqs, ents) %>% arrange(desc(cnts)))
  d
}

# Sim_ID == 12140 well illustrates the measures
dfy = dfb %>% filter(Sim_ID == 12140) #%>% group_by(Sim_ID)
de = entropy2D(dfy$Opinion1_Final, dfy$Opinion2_Final, 0.1)
hist(de, breaks = seq(0, 3, 0.1))







# Old graphs --------------------------------------------------------------



## Effects on size of components

ggplot(data, aes(x = size, fill = drawn, color = drawn)) +
  geom_density(alpha = 0.2) +
  labs(x = "Size of component as percentage of all agents", y = "Density",
       title = "Method of assigning boundary matters!",
       subtitle = "Random uniform distribution prefers components >80%\nassigning a constant value prefers components <10%.") +
  scale_x_continuous(breaks = seq(0, 100, 20), labels = function(x) paste0(x, "%")) +
  theme_minimal()
ggsave("Figs/Fig01-MethodOfAssigningBoundaryMatters.png", width = 6, height = 4)



data %>% mutate(neis = factor(neis)) %>%
  # filter(N_agents == 129) %>%
  ggplot(aes(x = size, fill = neis, color = neis)) +
  geom_density(alpha = 0.2) +
  labs(x = "Size of component as percentage of all agents", y = "Density",
       title = "Number of network neighbors matters!",
       subtitle = "The smaller neighborhood the more prefers smaller components <10%.") +
  scale_x_continuous(breaks = seq(0, 100, 20), labels = function(x) paste0(x, "%")) +
  theme_minimal()
ggsave("Figs/Fig02-SizeOfNeighborhoodMatters.png", width = 6, height = 4)



data %>% mutate(random_links = factor(random_links)) %>%
  # filter(neis == 4) %>%
  ggplot(aes(x = size, fill = random_links, color = random_links)) +
  geom_density(alpha = 0.2) +
  facet_grid(cols = vars(N_agents)) +
  labs(x = "Size of component as percentage of all agents", y = "Density",
       title = "Random re-wiring matters!",
       subtitle = "The smaller re-wiring more prefers smaller components <10%\n and number of agents in simulation intensifies this.") +
  scale_x_continuous(breaks = seq(20, 100, 40), labels = function(x) paste0(x, "%")) +
  theme_minimal()
ggsave("Figs/Fig03-RandomRewiringMatters.png", width = 6, height = 4)



##  Effects on length of simulation

data %>% filter(component_ID == 1) %>%  # Now we are interested in simulation as a whole so we have select just one compenent from whole simulation, since all components contain same information about whole simulation.
  ggplot(aes(x = step, fill = drawn, color = drawn)) +
  geom_density(alpha = 0.2) +
  labs(x = "Steps until the equilibrium", y = "Density",
       title = "Method of assigning boundary doesn't matter\nfor length of simulation.",
       #subtitle = "Random uniform distribution prefers components >80%\nassigning a constant value prefers components <10%."
  ) +
  scale_x_log10() + #breaks = seq(0, 100, 20), labels = function(x) paste0(x, "%")) +
  # scale_y_log10() +
  theme_minimal()
# NOTE: Since tehre is no effect we do not save the picture of graph.



data %>% filter(component_ID == 1 & step < 350) %>%  # Now we are interested in simulation as a whole so we have select just one compenent from whole simulation, since all components contain same information about whole simulation.
  mutate(neis = factor(neis)) %>%
  ggplot(aes(x = step, fill = neis, color = neis)) +
  geom_density(alpha = 0.2) +
  facet_grid(cols = vars(N_agents)) +
  labs(x = "Steps until the equilibrium", y = "Density",
       title = "Number of neighbors matters for length of simulation!",
       subtitle = "The more neighbors the shorter the simulation.\nIt's intensified by lower numbers of agents in simulation.",
       caption = "Note: for steps over 350 the neighborhood size doesn't matter,\nso we omit simulations that reach equilibrium after 350 steps."
  ) +
  # scale_x_log10() + #breaks = seq(0, 100, 20), labels = function(x) paste0(x, "%")) +
  # scale_y_log10() +
  theme_minimal()
ggsave("Figs/Fig04-SizeOfNeighborhoodMatters.png", width = 6, height = 4)



data %>% filter(component_ID == 1 & step < 350) %>%  # Now we are interested in simulation as a whole so we have select just one compenent from whole simulation, since all components contain same information about whole simulation.
  mutate(random_links = factor(random_links)) %>%
  ggplot(aes(x = step, fill = random_links, color = random_links)) +
  geom_density(alpha = 0.2) +
  facet_grid(cols = vars(N_agents)) +
  labs(x = "Steps until the equilibrium", y = "Density",
       title = "Random re-wiring matters for length of simulation!",
       subtitle = "The higher re-wiring the shorter the simulation.\nIt's intensified a bit by lower numbers of agents in simulation.",
       caption = "Note: for steps over 350 the neighborhood size doesn't matter,\nso we omit simulations that reach equilibrium after 350 steps."
  ) +
  # scale_x_log10() + #breaks = seq(0, 100, 20), labels = function(x) paste0(x, "%")) +
  # scale_y_log10() +
  theme_minimal()
ggsave("Figs/Fig05-RandomRewiringMatters.png", width = 6, height = 4)



## Effects on including simulation into main sample and reason of omitting

reason %>% mutate(neis = factor(neis)) %>%
  ggplot(aes(y = neis, fill = reason)) +
  geom_bar(position = "stack") +
  labs(title = "Size of neighborhood matters!",
       subtitle = "With larger neighborhood it's more likely\nthat simulation reaches equilibrium.\nBut constant is proportion of simulations\nwith all components smaller than 6.") +
  theme_minimal()
ggsave("Figs/Fig06-SizeOfNeighborhoodMatters.png", width = 6, height = 4)


reason %>% mutate(opinions = factor(opinions)) %>%
  ggplot(aes(y = opinions, fill = reason)) +
  geom_bar(position = "stack") +
  labs(title = "Number of opinions matters!",
       subtitle = "With more dimensions of opinion space is more likely\nthat all componets are of size smaller than 6.\nThe proportion of simulations not reaching equilibrium is more or less\nsame regardless the dimensions of opinion space.") +
  theme_minimal()
ggsave("Figs/Fig07-NumberOfOpinionsMatters.png", width = 6, height = 4)


reason %>% mutate(neis = factor(neis), opinions = factor(opinions)) %>%
  ggplot(aes(y = neis, fill = reason)) +
  facet_grid(vars(opinions)) +
  geom_bar(position = "fill") +
  scale_x_continuous(breaks = seq(0, 1, .20), labels = function(x) paste0(100*x, "%")) +
  labs(title = "Number of opinions and neighbors matter!\n('neis' on Y-axis, 'opinions' define panels)",
       subtitle = "Combination of larger neigborhood and less opinion dimensions supports\ninclusion of simulation into main file, i.e. simulation reaches equilibrium\nand at least some components are of size 6+. While more opinions\nfracture the public into groups smaller than 6, smaller neighborhood blocks\nreaching equilibrium.",
       x = "Proportion") +
  theme_minimal()
ggsave("Figs/Fig08-NumberOfOpinionsAndNeighborsMatter.png", width = 6, height = 4)