Newer
Older
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
print(first_index)
print("-> LAST INDEX:")
print(last_index)
print("-> STRUCTURE OF FIRST ROUND INDICES:")
print(str(sub_array_of_fri))
print("-> STRUCTURE OF SECOND ROUND INDICES:")
print(str(sub_array_of_sri))
print("-> STRUCTURE OF TRANSFORMED VARIABLE INDICES:")
print(str(tvi))
}
}
### # If the selectors are expressed after transformation
### } else {
###if (debug) {
###if (inner_dim %in% dims_to_check) {
###print("-> SELECTORS REQUESTED AFTER TRANSFORM.")
###}
###}
### if (goes_across_prime_meridian) {
### sub_array_of_indices <- c(sub_array_of_indices[[1]]:m,
### 1:sub_array_of_indices[[2]])
### }
### first_index <- min(unlist(sub_array_of_indices))
### last_index <- max(unlist(sub_array_of_indices))
### first_index_before_transform <- max(transform_indices(first_index, m, n) - beta, 1)
### last_index_before_transform <- min(transform_indices(last_index, m, n) + beta, n)
### sub_array_of_fri <- first_index_before_transform:last_index_before_transform
### n_of_extra_cells <- round(beta / n * m)
### if (is.list(sub_array_of_indices) && (length(sub_array_of_indices) > 1)) {
### sub_array_of_sri <- 1:(last_index - first_index + 1)
### if (is.null(tvi)) {
### tvi <- sub_array_of_sri + first_index - 1
### }
### } else {
### sub_array_of_sri <- sub_array_of_indices - first_index + 1
### if (is.null(tvi)) {
### tvi <- sub_array_of_indices
### }
### }
### sub_array_of_sri <- sub_array_of_sri + n_of_extra_cells
sri <- do.call('[[<-', c(list(x = sri), as.list(selector_store_position),
list(value = sub_array_of_sri)))
} else {
if (goes_across_prime_meridian) {
#sub_array_of_fri <- 1:n
sub_array_of_fri <- c(1:sub_array_of_indices[[2]], sub_array_of_indices[[1]]:n)
} else if (is.list(sub_array_of_indices)) {
sub_array_of_fri <- sub_array_of_indices[[1]]:sub_array_of_indices[[2]]
} else {
sub_array_of_fri <- sub_array_of_indices
}
}
if (!is.null(var_unorder_indices)) {
if (is.null(ordered_fri)) {
ordered_fri <- sub_array_of_fri
}
sub_array_of_fri <- var_unorder_indices[sub_array_of_fri]
}
fri <- do.call('[[<-', c(list(x = fri), as.list(selector_store_position),
list(value = sub_array_of_fri)))
if (!is.null(file_dim)) {
taken_chunks[selector_store_position[[file_dim]]] <- TRUE
} else {
taken_chunks <- TRUE
}
}
} else {
if (debug) {
if (inner_dim %in% dims_to_check) {
print("-> THE INNER DIMENSION GOES ACROSS A FILE DIMENSION.")
}
}
if (inner_dim %in% names(dim(sub_array_of_selectors))) {
if (is.null(var_with_selectors_name)) {
if (any(na.omit(unlist(sub_array_of_selectors)) < 1) ||
any(na.omit(unlist(sub_array_of_selectors)) > data_dims[inner_dim] * chunk_amount)) {
stop("Provided indices out of range for dimension '", inner_dim, "' ",
"for dataset '", dat[[i]][['name']], "' (accepted range: 1 to ",
data_dims[inner_dim] * chunk_amount, ").")
}
} else {
if (inner_dim %in% names(dim(sub_array_of_values))) {
inner_dim_pos_in_sub_array <- which(names(dim(sub_array_of_values)) == inner_dim)
if (inner_dim_pos_in_sub_array != 1) {
new_sub_array_order <- (1:length(dim(sub_array_of_values)))[-inner_dim_pos_in_sub_array]
new_sub_array_order <- c(inner_dim_pos_in_sub_array, new_sub_array_order)
sub_array_of_values <- .aperm2(sub_array_of_values, new_sub_array_order)
}
}
}
inner_dim_pos_in_sub_array <- which(names(dim(sub_array_of_selectors)) == inner_dim)
if (inner_dim_pos_in_sub_array != 1) {
new_sub_array_order <- (1:length(dim(sub_array_of_selectors)))[-inner_dim_pos_in_sub_array]
new_sub_array_order <- c(inner_dim_pos_in_sub_array, new_sub_array_order)
sub_array_of_selectors <- .aperm2(sub_array_of_selectors, new_sub_array_order)
}
sub_array_of_indices <- selector_checker(sub_array_of_selectors, sub_array_of_values,
tolerance = tolerance_params[[inner_dim]])
sub_array_of_indices <- sub_array_of_indices[chunk_indices(length(sub_array_of_indices),
chunks[[inner_dim]]['chunk'],
chunks[[inner_dim]]['n_chunks'],
inner_dim)]
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
sub_array_is_list <- FALSE
if (is.list(sub_array_of_indices)) {
sub_array_is_list <- TRUE
sub_array_of_indices <- unlist(sub_array_of_indices)
}
if (is.null(var_with_selectors_name)) {
indices_chunk <- floor((sub_array_of_indices - 1) / data_dims[inner_dim]) + 1
transformed_indices <- ((sub_array_of_indices - 1) %% data_dims[inner_dim]) + 1
} else {
indices_chunk <- floor((sub_array_of_indices - 1) / var_full_dims[inner_dim]) + 1
transformed_indices <- ((sub_array_of_indices - 1) %% var_full_dims[inner_dim]) + 1
}
if (sub_array_is_list) {
sub_array_of_indices <- as.list(sub_array_of_indices)
}
if (debug) {
if (inner_dim %in% dims_to_check) {
print("-> GOING TO ITERATE ALONG CHUNKS.")
}
}
for (chunk in 1:chunk_amount) {
if (!is.null(names(selector_store_position))) {
selector_store_position[file_dim] <- chunk
} else {
selector_store_position <- chunk
}
chunk_selectors <- transformed_indices[which(indices_chunk == chunk)]
sub_array_of_indices <- chunk_selectors
if (with_transform) {
# If the provided selectors are expressed in the world
# before transformation
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
first_index <- min(unlist(sub_array_of_indices))
last_index <- max(unlist(sub_array_of_indices))
sub_array_of_fri <- max(c(first_index - beta, 1)):min(c(last_index + beta, n))
sub_array_of_sri <- transform_indices(unlist(sub_array_of_indices) - first_index + 1, n, m)
if (is.list(sub_array_of_indices)) {
if (length(sub_array_of_sri) > 1) {
sub_array_of_sri <- sub_array_of_sri[[1]]:sub_array_of_sri[[2]]
}
}
##TODO: TRANSFORM SUBSET VARIABLE AS ABOVE, TO COMPUTE SRI
# If the selectors are expressed after transformation
} else {
first_index <- min(unlist(sub_array_of_indices))
last_index <- max(unlist(sub_array_of_indices))
first_index_before_transform <- max(transform_indices(first_index, m, n) - beta, 1)
last_index_before_transform <- min(transform_indices(last_index, m, n) + beta, n)
sub_array_of_fri <- first_index_before_transform:last_index_before_transform
if (is.list(sub_array_of_indices) && (length(sub_array_of_indices) > 1)) {
sub_array_of_sri <- 1:(last_index - first_index + 1) +
round(beta / n * m)
} else {
sub_array_of_sri <- sub_array_of_indices - first_index + 1 +
round(beta / n * m)
}
##TODO: FILL IN TVI
}
sri <- do.call('[[<-', c(list(x = sri), as.list(selector_store_position),
list(value = sub_array_of_sri)))
if (length(sub_array_of_sri) > 0) {
taken_chunks[chunk] <- TRUE
}
} else {
sub_array_of_fri <- sub_array_of_indices
if (length(sub_array_of_fri) > 0) {
taken_chunks[chunk] <- TRUE
}
}
if (!is.null(var_unorder_indices)) {
ordered_fri <- sub_array_of_fri
sub_array_of_fri <- var_unorder_indices[sub_array_of_fri]
}
fri <- do.call('[[<-', c(list(x = fri), as.list(selector_store_position),
list(value = sub_array_of_fri)))
}
if (debug) {
if (inner_dim %in% dims_to_check) {
print("-> FINISHED ITERATING ALONG CHUNKS")
}
}
} else {
stop("Provided array of indices for dimension '", inner_dim, "', ",
"which goes across the file dimension '", file_dim, "', but ",
"the provided array does not have the dimension '", inner_dim,
"', which is mandatory.")
}
}
}
}
if (debug) {
if (inner_dim %in% dims_to_check) {
print("-> PROCEEDING TO CROP VARIABLES")
}
}
#if ((length(selector_array) == 1) && (selector_array %in% c('all', 'first', 'last'))) {
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
#if (!is.null(var_with_selectors_name) || (is.null(var_with_selectors_name) && is.character(selector_array) &&
# (length(selector_array) == 1) && (selector_array %in% c('all', 'first', 'last')))) {
empty_chunks <- which(!taken_chunks)
if (length(empty_chunks) >= length(taken_chunks)) {
stop("Selectors do not match any of the possible values for the dimension '", inner_dim, "'.")
}
if (length(empty_chunks) > 0) {
# # Get the first group of chunks to remove, and remove them.
# # E.g., from c(1, 2, 4, 5, 6, 8, 9) remove only 1 and 2
# dist <- abs(rev(empty_chunks) - c(rev(empty_chunks)[1] - 1, head(rev(empty_chunks), length(rev(empty_chunks)) - 1)))
# if (all(dist == 1)) {
# start_chunks_to_remove <- NULL
# } else {
# first_chunk_to_remove <- tail(which(dist > 1), 1)
# start_chunks_to_remove <- rev(rev(empty_chunks)[first_chunk_to_remove:length(empty_chunks)])
# }
# # Get the last group of chunks to remove, and remove them.
# # E.g., from c(1, 2, 4, 5, 6, 8, 9) remove only 8 and 9
# dist <- abs(empty_chunks - c(empty_chunks[1] - 1, head(empty_chunks, length(empty_chunks) - 1)))
# if (all(dist == 1)) {
# first_chunk_to_remove <- 1
# } else {
# first_chunk_to_remove <- tail(which(dist > 1), 1)
# }
# end_chunks_to_remove <- empty_chunks[first_chunk_to_remove:length(empty_chunks)]
# chunks_to_keep <- which(!((1:length(taken_chunks)) %in% c(start_chunks_to_remove, end_chunks_to_remove)))
chunks_to_keep <- which(taken_chunks)
dims_to_crop[[file_dim]] <- c(dims_to_crop[[file_dim]], list(chunks_to_keep))
# found_indices <- Subset(found_indices, file_dim, chunks_to_keep)
# # Crop dataset variables file dims.
# for (picked_var in names(picked_vars[[i]])) {
# if (file_dim %in% names(dim(picked_vars[[i]][[picked_var]]))) {
# picked_vars[[i]][[picked_var]] <- Subset(picked_vars[[i]][[picked_var]], file_dim, chunks_to_keep)
# }
# }
}
#}
dat[[i]][['selectors']][[inner_dim]] <- list(fri = fri, sri = sri)
# Crop dataset variables inner dims.
# Crop common variables inner dims.
types_of_var_to_crop <- 'picked'
if (with_transform) {
types_of_var_to_crop <- c(types_of_var_to_crop, 'transformed')
}
if (!is.null(dim_reorder_params[[inner_dim]])) {
types_of_var_to_crop <- c(types_of_var_to_crop, 'reordered')
}
for (type_of_var_to_crop in types_of_var_to_crop) {
if (type_of_var_to_crop == 'transformed') {
if (is.null(tvi)) {
if (!is.null(dim_reorder_params[[inner_dim]])) {
crop_indices <- unique(unlist(ordered_sri))
} else {
crop_indices <- unique(unlist(sri))
}
} else {
crop_indices <- unique(unlist(tvi))
}
vars_to_crop <- transformed_vars[[i]]
common_vars_to_crop <- transformed_common_vars
} else if (type_of_var_to_crop == 'reordered') {
crop_indices <- unique(unlist(ordered_fri))
vars_to_crop <- picked_vars_ordered[[i]]
common_vars_to_crop <- picked_common_vars_ordered
} else {
crop_indices <- unique(unlist(fri))
vars_to_crop <- picked_vars[[i]]
common_vars_to_crop <- picked_common_vars
}
for (var_to_crop in names(vars_to_crop)) {
if (inner_dim %in% names(dim(vars_to_crop[[var_to_crop]]))) {
if (!is.null(crop_indices)) {
if (type_of_var_to_crop == 'transformed') {
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
vars_to_crop[[var_to_crop]] <- Subset(transformed_subset_var, inner_dim, crop_indices)
} else {
vars_to_crop[[var_to_crop]] <- Subset(vars_to_crop[[var_to_crop]], inner_dim, crop_indices)
}
} else {
vars_to_crop[[var_to_crop]] <- Subset(vars_to_crop[[var_to_crop]], inner_dim, crop_indices)
}
}
}
}
if (i == length(dat)) {
for (common_var_to_crop in names(common_vars_to_crop)) {
if (inner_dim %in% names(dim(common_vars_to_crop[[common_var_to_crop]]))) {
common_vars_to_crop[[common_var_to_crop]] <- Subset(common_vars_to_crop[[common_var_to_crop]], inner_dim, crop_indices)
}
}
}
if (type_of_var_to_crop == 'transformed') {
if (!is.null(vars_to_crop)) {
transformed_vars[[i]] <- vars_to_crop
}
if (i == length(dat)) {
transformed_common_vars <- common_vars_to_crop
}
} else if (type_of_var_to_crop == 'reordered') {
if (!is.null(vars_to_crop)) {
picked_vars_ordered[[i]] <- vars_to_crop
}
if (i == length(dat)) {
picked_common_vars_ordered <- common_vars_to_crop
}
} else {
if (!is.null(vars_to_crop)) {
picked_vars[[i]] <- vars_to_crop
}
if (i == length(dat)) {
picked_common_vars <- common_vars_to_crop
}
}
}
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
}
# After the selectors have been picked (using the original variables),
# the variables are transformed. At that point, the original selectors
# for the transformed variables are also kept in the variable original_selectors.
#print("L")
}
}
}
# if (!is.null(transformed_common_vars)) {
# picked_common_vars[names(transformed_common_vars)] <- transformed_common_vars
# }
# Remove the trailing chunks, if any.
for (file_dim in names(dims_to_crop)) {
# indices_to_keep <- min(sapply(dims_to_crop[[file_dim]], min)):max(sapply(dims_to_crop[[file_dim]], max))
## TODO: Merge indices in dims_to_crop with some advanced mechanism?
indices_to_keep <- unique(unlist(dims_to_crop[[file_dim]]))
array_of_files_to_load <- Subset(array_of_files_to_load, file_dim, indices_to_keep)
array_of_not_found_files <- Subset(array_of_not_found_files, file_dim, indices_to_keep)
for (i in 1:length(dat)) {
# Crop selectors
for (selector_dim in names(dat[[i]][['selectors']])) {
if (selector_dim == file_dim) {
for (j in 1:length(dat[[i]][['selectors']][[selector_dim]][['fri']])) {
dat[[i]][['selectors']][[selector_dim]][['fri']][[j]] <- dat[[i]][['selectors']][[selector_dim]][['fri']][[j]][indices_to_keep]
}
for (j in 1:length(dat[[i]][['selectors']][[selector_dim]][['sri']])) {
dat[[i]][['selectors']][[selector_dim]][['sri']][[j]] <- dat[[i]][['selectors']][[selector_dim]][['sri']][[j]][indices_to_keep]
}
}
if (file_dim %in% names(dim(dat[[i]][['selectors']][[selector_dim]][['fri']]))) {
dat[[i]][['selectors']][[selector_dim]][['fri']] <- Subset(dat[[i]][['selectors']][[selector_dim]][['fri']], file_dim, indices_to_keep)
dat[[i]][['selectors']][[selector_dim]][['sri']] <- Subset(dat[[i]][['selectors']][[selector_dim]][['sri']], file_dim, indices_to_keep)
}
}
# Crop dataset variables file dims.
for (picked_var in names(picked_vars[[i]])) {
if (file_dim %in% names(dim(picked_vars[[i]][[picked_var]]))) {
picked_vars[[i]][[picked_var]] <- Subset(picked_vars[[i]][[picked_var]], file_dim, indices_to_keep)
}
}
for (transformed_var in names(transformed_vars[[i]])) {
if (file_dim %in% names(dim(transformed_vars[[i]][[transformed_var]]))) {
transformed_vars[[i]][[transformed_var]] <- Subset(transformed_vars[[i]][[transformed_var]], file_dim, indices_to_keep)
}
}
}
# Crop common variables file dims.
for (picked_common_var in names(picked_common_vars)) {
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
if (file_dim %in% names(dim(picked_common_vars[[picked_common_var]]))) {
picked_common_vars[[picked_common_var]] <- Subset(picked_common_vars[[picked_common_var]], file_dim, indices_to_keep)
}
}
for (transformed_common_var in names(transformed_common_vars)) {
if (file_dim %in% names(dim(transformed_common_vars[[transformed_common_var]]))) {
transformed_common_vars[[transformed_common_var]] <- Subset(transformed_common_vars[[transformed_common_var]], file_dim, indices_to_keep)
}
}
}
# Calculate the size of the final array.
total_inner_dims <- NULL
for (i in 1:length(dat)) {
if (dataset_has_files[i]) {
inner_dims <- expected_inner_dims[[i]]
inner_dims <- sapply(inner_dims,
function(x) {
if (!all(sapply(dat[[i]][['selectors']][[x]][['sri']], is.null))) {
max(sapply(dat[[i]][['selectors']][[x]][['sri']], length))
} else {
if (length(var_params[[x]]) > 0) {
if (var_params[[x]] %in% names(transformed_vars[[i]])) {
length(transformed_vars[[i]][[var_params[[x]]]])
} else if (var_params[[x]] %in% names(transformed_common_vars)) {
length(transformed_common_vars[[var_params[[x]]]])
} else {
max(sapply(dat[[i]][['selectors']][[x]][['fri']], length))
}
} else {
max(sapply(dat[[i]][['selectors']][[x]][['fri']], length))
}
}
})
names(inner_dims) <- expected_inner_dims[[i]]
if (is.null(total_inner_dims)) {
total_inner_dims <- inner_dims
} else {
new_dims <- .MergeArrayDims(total_inner_dims, inner_dims)
total_inner_dims <- pmax(new_dims[[1]], new_dims[[2]])
}
}
}
new_dims <- .MergeArrayDims(dim(array_of_files_to_load), total_inner_dims)
final_dims <- pmax(new_dims[[1]], new_dims[[2]])[dim_names]
Nicolau Manubens
committed
# final_dims_fake is the vector of final dimensions after having merged the
# 'across' file dimensions with the respective 'across' inner dimensions, and
# after having broken into multiple dimensions those dimensions for which
# multidimensional selectors have been provided.
# final_dims will be used for collocation of data, whereas final_dims_fake
# will be used for shaping the final array to be returned to the user.
final_dims_fake <- final_dims
if (merge_across_dims) {
if (!is.null(inner_dims_across_files)) {
for (file_dim_across in names(inner_dims_across_files)) {
inner_dim_pos <- which(names(final_dims_fake) == inner_dims_across_files[[file_dim_across]])
Nicolau Manubens
committed
new_dims <- c()
if (inner_dim_pos > 1) {
new_dims <- c(new_dims, final_dims_fake[1:(inner_dim_pos - 1)])
}
new_dims <- c(new_dims, setNames(prod(final_dims_fake[c(inner_dim_pos, inner_dim_pos + 1)]),
inner_dims_across_files[[file_dim_across]]))
Nicolau Manubens
committed
if (inner_dim_pos + 1 < length(final_dims_fake)) {
new_dims <- c(new_dims, final_dims_fake[(inner_dim_pos + 2):length(final_dims_fake)])
}
final_dims_fake <- new_dims
}
}
}
Nicolau Manubens
committed
if (split_multiselected_dims) {
for (dim_param in 1:length(dim_params)) {
if (!is.null(dim(dim_params[[dim_param]]))) {
if (length(dim(dim_params[[dim_param]])) > 1) {
split_dims <- dim(dim_params[[dim_param]])
all_split_dims <- c(all_split_dims, setNames(list(split_dims),
names(dim_params)[dim_param]))
if (is.null(names(split_dims))) {
names(split_dims) <- paste0(names(dim_params)[dim_param],
1:length(split_dims))
Nicolau Manubens
committed
}
old_dim_pos <- which(names(final_dims_fake) == names(dim_params)[dim_param])
new_dims <- c()
if (old_dim_pos > 1) {
new_dims <- c(new_dims, final_dims_fake[1:(old_dim_pos - 1)])
}
new_dims <- c(new_dims, split_dims)
if (old_dim_pos < length(final_dims_fake)) {
new_dims <- c(new_dims, final_dims_fake[(old_dim_pos + 1):length(final_dims_fake)])
}
Nicolau Manubens
committed
}
}
}
}
if (!silent) {
.message("Detected dimension sizes:")
Nicolau Manubens
committed
longest_dim_len <- max(sapply(names(final_dims_fake), nchar))
longest_size_len <- max(sapply(paste0(final_dims_fake, ''), nchar))
sapply(names(final_dims_fake),
function(x) {
message(paste0("* ", paste(rep(' ', longest_dim_len - nchar(x)), collapse = ''),
Nicolau Manubens
committed
x, ": ", paste(rep(' ', longest_size_len - nchar(paste0(final_dims_fake[x], ''))), collapse = ''),
final_dims_fake[x]))
Nicolau Manubens
committed
bytes <- prod(c(final_dims_fake, 8))
dim_sizes <- paste(final_dims_fake, collapse = ' x ')
if (retrieve) {
.message(paste("Total size of requested data:"))
} else {
.message(paste("Total size of involved data:"))
}
.message(paste(dim_sizes, " x 8 bytes =",
format(structure(bytes, class = "object_size"), units = "auto")),
indent = 2)
}
# The following several lines will only be run if retrieve = TRUE
if (retrieve) {
########## CREATING THE SHARED MATRIX AND DISPATCHING WORK PIECES ###########
# TODO: try performance of storing all in cols instead of rows
# Create the shared memory array, and a pointer to it, to be sent
# to the work pieces.
data_array <- big.matrix(nrow = prod(final_dims), ncol = 1)
shared_matrix_pointer <- describe(data_array)
if (is.null(num_procs)) {
# Creating a shared tmp folder to store metadata from each chunk
array_of_metadata_flags <- array(FALSE, dim = dim(array_of_files_to_load))
if (!is.null(metadata_dims)) {
metadata_indices_to_load <- as.list(rep(1, length(dim(array_of_files_to_load))))
names(metadata_indices_to_load) <- names(dim(array_of_files_to_load))
metadata_indices_to_load[metadata_dims] <- as.list(rep(TRUE, length(metadata_dims)))
array_of_metadata_flags <- do.call('[<-', c(list(array_of_metadata_flags), metadata_indices_to_load,
list(value = rep(TRUE, prod(dim(array_of_files_to_load)[metadata_dims])))))
}
metadata_file_counter <- 0
metadata_folder <- tempfile('metadata')
dir.create(metadata_folder)
# Build the work pieces, each with:
# - file path
# - total size (dims) of store array
# - start position in store array
# - file selectors (to provide extra info. useful e.g. to select variable)
# - indices to take from file
work_pieces <- list()
for (i in 1:length(dat)) {
if (dataset_has_files[i]) {
selectors <- dat[[i]][['selectors']]
file_dims <- found_file_dims[[i]]
inner_dims <- expected_inner_dims[[i]]
sub_array_dims <- final_dims[file_dims]
sub_array_dims[found_pattern_dim] <- 1
sub_array_of_files_to_load <- array(1:prod(sub_array_dims),
dim = sub_array_dims)
names(dim(sub_array_of_files_to_load)) <- names(sub_array_dims)
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
# Detect which of the dimensions of the dataset go across files.
file_dim_across_files <- lapply(inner_dims,
function(x) {
dim_across <- sapply(inner_dims_across_files, function(y) x %in% y)
if (any(dim_across)) {
names(inner_dims_across_files)[which(dim_across)[1]]
} else {
NULL
}
})
names(file_dim_across_files) <- inner_dims
j <- 1
while (j <= prod(sub_array_dims)) {
# Work out file path.
file_to_load_sub_indices <- which(sub_array_of_files_to_load == j, arr.ind = TRUE)[1, ]
names(file_to_load_sub_indices) <- names(sub_array_dims)
file_to_load_sub_indices[found_pattern_dim] <- i
big_dims <- rep(1, length(dim(array_of_files_to_load)))
names(big_dims) <- names(dim(array_of_files_to_load))
file_to_load_indices <- .MergeArrayDims(file_to_load_sub_indices, big_dims)[[1]]
file_to_load <- do.call('[[', c(list(array_of_files_to_load),
as.list(file_to_load_indices)))
not_found_file <- do.call('[[', c(list(array_of_not_found_files),
as.list(file_to_load_indices)))
load_file_metadata <- do.call('[', c(list(array_of_metadata_flags),
as.list(file_to_load_indices)))
if (load_file_metadata) {
metadata_file_counter <- metadata_file_counter + 1
}
if (!is.na(file_to_load) && !not_found_file) {
# Work out indices to take
first_round_indices <- lapply(inner_dims,
function (x) {
if (is.null(file_dim_across_files[[x]])) {
selectors[[x]][['fri']][[1]]
} else {
which_chunk <- file_to_load_sub_indices[file_dim_across_files[[x]]]
selectors[[x]][['fri']][[which_chunk]]
}
})
names(first_round_indices) <- inner_dims
second_round_indices <- lapply(inner_dims,
function (x) {
if (is.null(file_dim_across_files[[x]])) {
selectors[[x]][['sri']][[1]]
} else {
which_chunk <- file_to_load_sub_indices[file_dim_across_files[[x]]]
selectors[[x]][['sri']][[which_chunk]]
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
}
})
if (debug) {
print("-> BUILDING A WORK PIECE")
#print(str(selectors))
}
names(second_round_indices) <- inner_dims
if (!any(sapply(first_round_indices, length) == 0)) {
work_piece <- list()
work_piece[['first_round_indices']] <- first_round_indices
work_piece[['second_round_indices']] <- second_round_indices
work_piece[['file_indices_in_array_of_files']] <- file_to_load_indices
work_piece[['file_path']] <- file_to_load
work_piece[['store_dims']] <- final_dims
# Work out store position
store_position <- final_dims
store_position[names(file_to_load_indices)] <- file_to_load_indices
store_position[inner_dims] <- rep(1, length(inner_dims))
work_piece[['store_position']] <- store_position
# Work out file selectors
file_selectors <- sapply(file_dims,
function (x) {
vector_to_pick <- 1
if (x %in% names(depending_file_dims)) {
vector_to_pick <- file_to_load_indices[depending_file_dims[[x]]]
}
selectors[file_dims][[x]][[vector_to_pick]][file_to_load_indices[x]]
})
names(file_selectors) <- file_dims
work_piece[['file_selectors']] <- file_selectors
# Send variables for transformation
if (!is.null(transform) && (length(transform_vars) > 0)) {
vars_to_transform <- NULL
picked_vars_to_transform <- which(names(picked_vars[[i]]) %in% transform_vars)
if (length(picked_vars_to_transform) > 0) {
picked_vars_to_transform <- names(picked_vars[[i]])[picked_vars_to_transform]
vars_to_transform <- c(vars_to_transform, picked_vars[[i]][picked_vars_to_transform])
if (any(picked_vars_to_transform %in% names(picked_vars_ordered[[i]]))) {
picked_vars_ordered_to_transform <- picked_vars_to_transform[which(picked_vars_to_transform %in% names(picked_vars_ordered[[i]]))]
vars_to_transform[picked_vars_ordered_to_transform] <- picked_vars_ordered[[i]][picked_vars_ordered_to_transform]
}
}
picked_common_vars_to_transform <- which(names(picked_common_vars) %in% transform_vars)
if (length(picked_common_vars_to_transform) > 0) {
picked_common_vars_to_transform <- names(picked_common_vars)[picked_common_vars_to_transform]
vars_to_transform <- c(vars_to_transform, picked_common_vars[picked_common_vars_to_transform])
if (any(picked_common_vars_to_transform %in% names(picked_common_vars_ordered))) {
picked_common_vars_ordered_to_transform <- picked_common_vars_to_transform[which(picked_common_vars_to_transform %in% names(picked_common_vars_ordered))]
vars_to_transform[picked_common_vars_ordered_to_transform] <- picked_common_vars_ordered[picked_common_vars_ordered_to_transform]
}
}
work_piece[['vars_to_transform']] <- vars_to_transform
}
# Send flag to load metadata
if (load_file_metadata) {
work_piece[['save_metadata_in']] <- paste0(metadata_folder, '/', metadata_file_counter)
}
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
work_pieces <- c(work_pieces, list(work_piece))
}
}
j <- j + 1
}
}
}
#print("N")
if (debug) {
print("-> WORK PIECES BUILT")
}
# Calculate the progress %s that will be displayed and assign them to
# the appropriate work pieces.
if (length(work_pieces) / num_procs >= 2 && !silent) {
if (length(work_pieces) / num_procs < 10) {
amount <- 100 / ceiling(length(work_pieces) / num_procs)
reps <- ceiling(length(work_pieces) / num_procs)
} else {
amount <- 10
reps <- 10
}
progress_steps <- rep(amount, reps)
if (length(work_pieces) < (reps + 1)) {
selected_pieces <- length(work_pieces)
progress_steps <- c(sum(head(progress_steps, reps)),
tail(progress_steps, reps))
} else {
selected_pieces <- round(seq(1, length(work_pieces),
length.out = reps + 1))[-1]
}
progress_steps <- paste0(' + ', round(progress_steps, 2), '%')
progress_message <- 'Progress: 0%'
} else {
progress_message <- ''
selected_pieces <- NULL
}
piece_counter <- 1
step_counter <- 1
work_pieces <- lapply(work_pieces,
function (x) {
if (piece_counter %in% selected_pieces) {
wp <- c(x, list(progress_amount = progress_steps[step_counter]))
step_counter <<- step_counter + 1
} else {
wp <- x
}
piece_counter <<- piece_counter + 1
wp
})
if (!silent) {
.message("If the size of the requested data is close to or above the free shared RAM memory, R may crash.")
.message("If the size of the requested data is close to or above the half of the free RAM memory, R may crash.")
.message(paste0("Will now proceed to read and process ", length(work_pieces), " data files:"))
if (length(work_pieces) < 30) {
lapply(work_pieces, function (x) .message(x[['file_path']], indent = 2))
} else {
.message("The list of files is long. You can check it after Start() finishes in the output '$Files'.", indent = 2, exdent = 5)
}
}
# Build the cluster of processes that will do the work and dispatch work pieces.
# The function .LoadDataFile is applied to each work piece. This function will
# open the data file, regrid if needed, subset, apply the mask,
# compute and apply the weights if needed,
# disable extreme values and store in the shared memory matrix.
#print("O")
if (!silent) {
.message("Loading... This may take several minutes...")
if (progress_message != '') {
.message(progress_message, appendLF = FALSE)
}
}
if (num_procs == 1) {
found_files <- lapply(work_pieces, .LoadDataFile,
shared_matrix_pointer = shared_matrix_pointer,
file_data_reader = file_data_reader,
transform = transform,
transform_params = transform_params,
silent = silent, debug = debug)
} else {
cluster <- makeCluster(num_procs, outfile = "")
# Send the heavy work to the workers
work_errors <- try({
found_files <- clusterApplyLB(cluster, work_pieces, .LoadDataFile,
shared_matrix_pointer = shared_matrix_pointer,
file_data_reader = file_data_reader,
transform = transform,
transform_params = transform_params,
silent = silent, debug = debug)
})
stopCluster(cluster)
}
if (!silent) {
if (progress_message != '') {
.message("\n", tag = '')
}
}
#print("P")
Nicolau Manubens
committed
data_array <- array(bigmemory::as.matrix(data_array), dim = final_dims_fake)
# Load metadata and remove the metadata folder
if (!is.null(metadata_dims)) {
loaded_metadata_files <- list.files(metadata_folder)
loaded_metadata <- lapply(paste0(metadata_folder, '/', loaded_metadata_files), readRDS)
unlink(metadata_folder, recursive = TRUE)
return_metadata <- vector('list', length = prod(dim(array_of_metadata_flags)[metadata_dims]))
return_metadata[as.numeric(loaded_metadata_files)] <- loaded_metadata
dim(return_metadata) <- dim(array_of_metadata_flags[metadata_dims])
attr(data_array, 'Variables') <- return_metadata
# TODO: Try to infer data type from loaded_metadata
# as.integer(data_array)
}
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
failed_pieces <- work_pieces[which(unlist(found_files))]
for (failed_piece in failed_pieces) {
array_of_not_found_files <- do.call('[<-',
c(list(array_of_not_found_files),
as.list(failed_piece[['file_indices_in_array_of_files']]),
list(value = TRUE)))
}
if (any(array_of_not_found_files)) {
for (i in 1:prod(dim(array_of_files_to_load))) {
if (is.na(array_of_not_found_files[i])) {
array_of_files_to_load[i] <- NA
} else {
if (array_of_not_found_files[i]) {
array_of_not_found_files[i] <- array_of_files_to_load[i]
array_of_files_to_load[i] <- NA
} else {
array_of_not_found_files[i] <- NA
}
}
}
} else {
array_of_not_found_files <- NULL
}
# Replace the vars and common vars by the transformed vars and common vars
for (i in 1:length(dat)) {
if (length(names(transformed_vars[[i]])) > 0) {
picked_vars[[i]][names(transformed_vars[[i]])] <- transformed_vars[[i]]
} else if (length(names(picked_vars_ordered[[i]])) > 0) {
picked_vars[[i]][names(picked_vars_ordered[[i]])] <- picked_vars_ordered[[i]]
}
}
if (length(names(transformed_common_vars)) > 0) {
picked_common_vars[names(transformed_common_vars)] <- transformed_common_vars
} else if (length(names(picked_common_vars_ordered)) > 0) {
picked_common_vars[names(picked_common_vars_ordered)] <- picked_common_vars_ordered
}
if (debug) {
print("-> THE TRANSFORMED VARS:")
print(str(transformed_vars))
print("-> THE PICKED VARS:")
print(str(picked_vars))
}
file_selectors <- NULL
for (i in 1:length(dat)) {
file_selectors[[dat[[i]][['name']]]] <- dat[[i]][['selectors']][which(names(dat[[i]][['selectors']]) %in% found_file_dims[[i]])]
}
if (retrieve) {
if (!silent) {
.message("Successfully retrieved data.")
}
var_backup <- attr(data_array, 'Variables')[[1]]
attr(data_array, 'Variables') <- NULL
attributes(data_array) <- c(attributes(data_array),
list(Variables = c(list(common = c(picked_common_vars, var_backup)),
picked_vars),
Files = array_of_files_to_load,
NotFoundFiles = array_of_not_found_files,
FileSelectors = file_selectors)
)
attr(data_array, 'class') <- c('startR_cube', attr(data_array, 'class'))
data_array
} else {
if (!silent) {
.message("Successfully discovered data dimensions.")
}
start_call <- match.call()
start_call[['retrieve']] <- TRUE
attributes(start_call) <- c(attributes(start_call),
Nicolau Manubens
committed
list(Dimensions = final_dims_fake,
Variables = c(list(common = picked_common_vars), picked_vars),
ExpectedFiles = array_of_files_to_load,
FileSelectors = file_selectors,
MergedDims = if (merge_across_dims) {
inner_dims_across_files
} else {
NULL
},
SplitDims = if (split_multiselected_dims) {
all_split_dims
} else {
NULL
})
)
attr(start_call, 'class') <- c('startR_header', attr(start_call, 'class'))
start_call
}
# This function is the responsible for loading the data of each work
# piece.
.LoadDataFile <- function(work_piece, shared_matrix_pointer,
file_data_reader, synonims,
transform, transform_params,
silent = FALSE, debug = FALSE) {
# suppressPackageStartupMessages({library(bigmemory)})
### TODO: Specify dependencies as parameter
# suppressPackageStartupMessages({library(ncdf4)})
#print("1")
store_indices <- as.list(work_piece[['store_position']])
first_round_indices <- work_piece[['first_round_indices']]
second_round_indices <- work_piece[['second_round_indices']]
#print("2")
file_to_open <- work_piece[['file_path']]
sub_array <- file_data_reader(file_to_open, NULL,
work_piece[['file_selectors']],
first_round_indices, synonims)
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
if (debug) {
if (all(unlist(store_indices[1:6]) == 1)) {
print("-> LOADING A WORK PIECE")
print("-> STRUCTURE OF READ UNTRANSFORMED DATA:")
print(str(sub_array))
print("-> STRUCTURE OF VARIABLES TO TRANSFORM:")
print(str(work_piece[['vars_to_transform']]))
print("-> COMMON ARRAY DIMENSIONS:")
print(str(work_piece[['store_dims']]))
}
}
if (!is.null(sub_array)) {
# Apply data transformation once we have the data arrays.
if (!is.null(transform)) {
if (debug) {
if (all(unlist(store_indices[1:6]) == 1)) {
print("-> PROCEEDING TO TRANSFORM ARRAY")
print("-> DIMENSIONS OF ARRAY RIGHT BEFORE TRANSFORMING:")
print(dim(sub_array))
}
}
sub_array <- do.call(transform, c(list(data_array = sub_array,
variables = work_piece[['vars_to_transform']],
file_selectors = work_piece[['file_selectors']]),
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
transform_params))
if (debug) {
if (all(unlist(store_indices[1:6]) == 1)) {
print("-> STRUCTURE OF ARRAY AND VARIABLES RIGHT AFTER TRANSFORMING:")
print(str(sub_array))
print("-> DIMENSIONS OF ARRAY RIGHT AFTER TRANSFORMING:")
print(dim(sub_array$data_array))
}
}
sub_array <- sub_array$data_array
# Subset with second round of indices
dims_to_crop <- which(!sapply(second_round_indices, is.null))
if (length(dims_to_crop) > 0) {
dimnames_to_crop <- names(second_round_indices)[dims_to_crop]
sub_array <- Subset(sub_array, dimnames_to_crop,
second_round_indices[dimnames_to_crop])
}
if (debug) {
if (all(unlist(store_indices[1:6]) == 1)) {
print("-> STRUCTURE OF ARRAY AND VARIABLES RIGHT AFTER SUBSETTING WITH 2nd ROUND INDICES:")
print(str(sub_array))
}
}
}
metadata <- attr(sub_array, 'variables')
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
store_indices <- lapply(names(store_indices),
function (x) {
if (!(x %in% names(first_round_indices))) {
store_indices[[x]]
} else if (is.null(second_round_indices[[x]])) {
1:dim(sub_array)[x]
} else {
if (is.numeric(second_round_indices[[x]])) {
## TODO: Review carefully this line. Inner indices are all
## aligned to the left-most positions. If dataset A has longitudes
## 1, 2, 3, 4 but dataset B has only longitudes 3 and 4, then
## they will be stored as follows:
## 1, 2, 3, 4
## 3, 4, NA, NA
##x - min(x) + 1
1:length(second_round_indices[[x]])
} else {
1:length(second_round_indices[[x]])
}
}
})
if (debug) {
print("-> STRUCTURE OF FIRST ROUND INDICES FOR THIS WORK PIECE:")
print(str(first_round_indices))
print("-> STRUCTURE OF SECOND ROUND INDICES FOR THIS WORK PIECE:")
print(str(second_round_indices))
print("-> STRUCTURE OF STORE INDICES FOR THIS WORK PIECE:")
print(str(store_indices))
}
}
# Converting array indices to vector indices
store_dims <- work_piece[['store_dims']]
store_indices <- do.call('expand.grid', store_indices)
### Given a matrix where each row is a set of array indices of an element
### the vector indices are computed
weights <- sapply(1:length(store_dims),
function(i) prod(c(1, store_dims)[1:i]))
matrix_indices <- 1 + colSums(t(store_indices - 1) * weights)
###
data_array <- bigmemory::attach.big.matrix(shared_matrix_pointer)
rm(data_array)
gc()
if (!is.null(work_piece[['save_metadata_in']])) {
saveRDS(metadata, file = work_piece[['save_metadata_in']])
}
}
if (!is.null(work_piece[['progress_amount']]) && !silent) {
message(work_piece[['progress_amount']], appendLF = FALSE)
}
is.null(sub_array)
}