-
Notifications
You must be signed in to change notification settings - Fork 64
/
stats.rs
618 lines (581 loc) · 18.6 KB
/
stats.rs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
use fnv::{FnvBuildHasher, FnvHashSet};
use indexmap::IndexMap;
use itertools::Itertools;
use num::ToPrimitive;
use std::{cmp::Ordering, collections::BTreeMap, num::NonZeroU64};
use tangram_finite::Finite;
use tangram_progress_counter::ProgressCounter;
use tangram_table::prelude::*;
use tangram_text::{NGram, NGramType, Tokenizer};
use tangram_zip::zip;
/// This struct contains settings used to compute stats.
#[derive(Clone, Debug)]
pub struct StatsSettings {
/// This is the maximum number of unique numeric values to store in the histogram.
pub number_histogram_max_size: usize,
/// This is the maximum number of ngrams to track for text columns.
pub ngrams_max_count: usize,
/// This setting specifies which ngram types should be computed.
pub ngram_types: FnvHashSet<NGramType>,
}
impl Default for StatsSettings {
fn default() -> StatsSettings {
StatsSettings {
number_histogram_max_size: 100,
ngrams_max_count: 20_000,
ngram_types: vec![NGramType::Unigram, NGramType::Bigram]
.into_iter()
.collect(),
}
}
}
#[derive(Clone, Debug)]
pub struct TextColumnStatsSettings {
pub tokenizer: Tokenizer,
}
/// This struct holds column stats.
#[derive(Clone, Debug)]
pub struct Stats(pub Vec<ColumnStats>);
/// This is an enum describing the different types of stats where the type matches the type of the source column.
#[derive(Clone, Debug)]
pub enum ColumnStats {
Unknown(UnknownColumnStats),
Number(NumberColumnStats),
Enum(EnumColumnStats),
Text(TextColumnStats),
}
/// This struct contains stats for unknown columns.
#[derive(Clone, Debug)]
pub struct UnknownColumnStats {
/// This is the name of the column.
pub column_name: String,
pub count: usize,
pub invalid_count: usize,
}
/// This struct contains stats for number columns.
#[derive(Clone, Debug)]
pub struct NumberColumnStats {
/// This is the name of the column.
pub column_name: String,
/// The total number of values.
pub count: usize,
/// The total number of valid values.
pub valid_count: usize,
/// This is the total number of invalid values. Invalid values are values that fail to parse as finite f32.
pub invalid_count: usize,
/// This stores counts for each unique value.
pub histogram: BTreeMap<Finite<f32>, usize>,
}
/// This struct contains stats for enum columns.
#[derive(Clone, Debug)]
pub struct EnumColumnStats {
/// This is the name of the column.
pub column_name: String,
/// This is the total number of values.
pub count: usize,
/// The enum variants.
pub variants: Vec<String>,
/// This is the total number of valid values.
pub valid_count: usize,
/// This is the total number of invalid values.
pub invalid_count: usize,
/// This is the histogram.
pub histogram: Vec<usize>,
}
/// This struct contains stats for text columns.
#[derive(Clone, Debug)]
pub struct TextColumnStats {
/// This is the name of the column.
pub column_name: String,
/// The total number of values.
pub row_count: usize,
/// This is the tokenizer used to split the text into tokens.
pub tokenizer: Tokenizer,
/// These are the types of ngrams to collect.
pub ngram_types: FnvHashSet<NGramType>,
/// These are stats collected on ngrams.
pub ngrams: IndexMap<NGram, TextColumnStatsNGramEntry, FnvBuildHasher>,
}
#[derive(Clone, Debug, Default)]
pub struct TextColumnStatsNGramEntry {
pub row_count: usize,
pub occurrence_count: usize,
}
pub struct StatsOutput(pub Vec<ColumnStatsOutput>);
/// This enum describes the different types of column stats.
#[derive(Debug)]
pub enum ColumnStatsOutput {
Unknown(UnknownColumnStatsOutput),
Number(NumberColumnStatsOutput),
Enum(EnumColumnStatsOutput),
Text(TextColumnStatsOutput),
}
impl ColumnStatsOutput {
/// Return the name of the source column.
pub fn column_name(&self) -> &str {
match self {
ColumnStatsOutput::Unknown(value) => &value.column_name,
ColumnStatsOutput::Number(value) => &value.column_name,
ColumnStatsOutput::Enum(value) => &value.column_name,
ColumnStatsOutput::Text(value) => &value.column_name,
}
}
}
/// This struct contains stats for unknown columns.
#[derive(Debug)]
pub struct UnknownColumnStatsOutput {
/// This is the name of the column as it appears in the csv.
pub column_name: String,
/// This is the total number of examples that these stats were computed on.
pub count: usize,
}
/// This struct contains stats for number columns.
#[derive(Debug)]
pub struct NumberColumnStatsOutput {
/// This is the name of the column as it appears in the csv.
pub column_name: String,
/// This is the total number of examples that these stats were computed on.
pub count: usize,
/// This is a histogram mapping unique values to their counts. It is `None` if the number of unique values exceeds [`number_histogram_max_size`](StatsSettings#number_histogram_max_size).
pub histogram: Option<Vec<(Finite<f32>, usize)>>,
/// This is the total number of unique values.
pub unique_count: usize,
/// This is the max of the values in the column.
pub max: f32,
/// This is the mean of the values in the column.
pub mean: f32,
/// This is the min of the values in the column.
pub min: f32,
/// This is the total number of invalid values. Invalid values are values that fail to parse as floating point numbers.
pub invalid_count: usize,
/// This is the variance of the values in the column.
pub variance: f32,
/// This is the standard deviation of the values in the column. It is equal to the square root of the variance.
pub std: f32,
/// This is the p25, or 25th-percentile value in the column.
pub p25: f32,
/// This is the p50, or 50th-percentile value in the column, i.e. the median.
pub p50: f32,
/// This is the p75, or 75th-percentile value in the column.
pub p75: f32,
}
/// This struct contains stats for enum columns.
#[derive(Debug)]
pub struct EnumColumnStatsOutput {
/// This is the name of the column as it appears in the csv.
pub column_name: String,
/// This is the total number of examples that these stats were computed on.
pub count: u64,
/// This is a histogram mapping unique variants of the enum to the total count of occurrences of the variant in the dataset.
pub histogram: Vec<(String, usize)>,
/// This is the total number of values in the dataset that are invalid. A value is invalid if it is not one of the enum's variants.
pub invalid_count: usize,
/// This is the total number of unique values, excluding invalid values.
pub unique_count: usize,
}
/// This struct contains stats for text columns.
#[derive(Debug)]
pub struct TextColumnStatsOutput {
/// This is the name of the column.
pub column_name: String,
/// This is the number of rows that these stats were computed for.
pub row_count: u64,
/// This is the tokenizer that was used to separate values into tokens.
pub tokenizer: Tokenizer,
/// This is the list of ngram types.
pub ngram_types: FnvHashSet<NGramType>,
/// This is the number of unique ngrams encountered.
pub ngrams_count: usize,
/// This contains stats for up to `stats_settings.ngrams_max_count` ngrams with the highest `entry.row_count`s.
pub top_ngrams: IndexMap<NGram, TextColumnStatsOutputTopNGramsEntry, FnvBuildHasher>,
}
/// This struct contains stats for individual ngrams.
#[derive(Clone, Debug)]
pub struct TextColumnStatsOutputTopNGramsEntry {
/// This is the number of rows that contain at least one occurrence of this ngram.
pub row_count: usize,
/// This is the number of occurrences of this ngram across all rows.
pub occurrence_count: usize,
/// This is the inverse document frequency of this ngram. [Learn more](https://en.wikipedia.org/wiki/Tf%E2%80%93idf).
pub idf: f32,
}
impl Stats {
pub fn compute(
table: &TableView,
settings: &StatsSettings,
handle_progress_event: &mut dyn FnMut(ProgressCounter),
) -> Stats {
let progress_total = table.ncols() as u64 * table.nrows() as u64;
let progress_counter = ProgressCounter::new(progress_total);
handle_progress_event(progress_counter.clone());
let progress = &|progress| progress_counter.inc(progress);
let column_stats = table
.columns()
.iter()
.map(|column| ColumnStats::compute(column.view(), settings, progress))
.collect();
Stats(column_stats)
}
pub fn merge(self, other: Stats) -> Stats {
let column_stats: Vec<ColumnStats> =
zip!(self.0, other.0).map(|(a, b)| a.merge(b)).collect();
Stats(column_stats)
}
pub fn finalize(self, settings: &StatsSettings) -> StatsOutput {
let column_stats = self
.0
.into_iter()
.map(|column_stats| column_stats.finalize(settings))
.collect();
StatsOutput(column_stats)
}
}
impl ColumnStats {
fn compute(
column: TableColumnView,
settings: &StatsSettings,
progress: impl Fn(u64),
) -> ColumnStats {
match column {
TableColumnView::Unknown(column) => {
progress(column.len() as u64);
ColumnStats::Unknown(UnknownColumnStats {
column_name: column.name().unwrap().to_owned(),
count: column.len(),
invalid_count: column.len(),
})
}
TableColumnView::Number(column) => ColumnStats::Number(NumberColumnStats::compute(
column.view(),
settings,
progress,
)),
TableColumnView::Enum(column) => {
ColumnStats::Enum(EnumColumnStats::compute(column, settings, progress))
}
TableColumnView::Text(column) => {
ColumnStats::Text(TextColumnStats::compute(column, settings, progress))
}
}
}
fn merge(self, other: ColumnStats) -> ColumnStats {
match (self, other) {
(ColumnStats::Unknown(a), ColumnStats::Unknown(b)) => {
ColumnStats::Unknown(UnknownColumnStats {
column_name: a.column_name.clone(),
count: a.count + b.count,
invalid_count: a.invalid_count + b.invalid_count,
})
}
(ColumnStats::Number(a), ColumnStats::Number(b)) => ColumnStats::Number(a.merge(b)),
(ColumnStats::Enum(a), ColumnStats::Enum(b)) => ColumnStats::Enum(a.merge(b)),
(ColumnStats::Text(a), ColumnStats::Text(b)) => ColumnStats::Text(a.merge(b)),
_ => unreachable!(),
}
}
fn finalize(self, settings: &StatsSettings) -> ColumnStatsOutput {
match self {
ColumnStats::Unknown(column_stats_output) => {
ColumnStatsOutput::Unknown(UnknownColumnStatsOutput {
column_name: column_stats_output.column_name,
count: column_stats_output.count,
})
}
ColumnStats::Number(column_stats_output) => {
ColumnStatsOutput::Number(column_stats_output.finalize(settings))
}
ColumnStats::Enum(column_stats_output) => {
ColumnStatsOutput::Enum(column_stats_output.finalize(settings))
}
ColumnStats::Text(column_stats_output) => {
ColumnStatsOutput::Text(column_stats_output.finalize(settings))
}
}
}
}
impl NumberColumnStats {
fn compute(
column: NumberTableColumnView,
_settings: &StatsSettings,
progress: impl Fn(u64),
) -> NumberColumnStats {
let mut stats = NumberColumnStats {
column_name: column.name().unwrap().to_owned(),
count: column.len(),
histogram: BTreeMap::new(),
invalid_count: 0,
valid_count: 0,
};
for value in column.iter() {
// If the value parses as a finite f32, add it to the histogram. Otherwise, increment the invalid count.
if let Ok(value) = <Finite<f32>>::new(*value) {
*stats.histogram.entry(value).or_insert(0) += 1;
stats.valid_count += 1;
} else {
stats.invalid_count += 1;
}
progress(1);
}
stats
}
fn merge(mut self, other: NumberColumnStats) -> NumberColumnStats {
for (value, count) in other.histogram.iter() {
*self.histogram.entry(*value).or_insert(0) += count;
}
self.count += other.count;
self.invalid_count += other.invalid_count;
self.valid_count += other.valid_count;
self
}
fn finalize(self, settings: &StatsSettings) -> NumberColumnStatsOutput {
let unique_values_count = self.histogram.len();
let invalid_count = self.invalid_count;
let histogram = if self.histogram.len() <= settings.number_histogram_max_size {
Some(self.histogram.iter().map(|(k, v)| (*k, *v)).collect())
} else {
None
};
let min = self.histogram.iter().next().unwrap().0.get();
let max = self.histogram.iter().next_back().unwrap().0.get();
let total_values_count = self.valid_count.to_f32().unwrap();
let quantiles: Vec<f32> = vec![0.25, 0.50, 0.75];
// Find the index of each quantile given the total number of values in the dataset.
let quantile_indexes: Vec<usize> = quantiles
.iter()
.map(|q| ((total_values_count - 1.0) * q).trunc().to_usize().unwrap())
.collect();
// This is the fractiononal part of the index used to interpolate values if the index is not an integer value.
let quantile_fracts: Vec<f32> = quantiles
.iter()
.map(|q| ((total_values_count - 1.0) * q).fract())
.collect();
let mut quantiles: Vec<Option<f32>> = vec![None; quantiles.len()];
let mut current_count: usize = 0;
let mut mean = 0.0;
let mut m2 = 0.0;
let mut iter = self.histogram.iter().peekable();
while let Some((value, count)) = iter.next() {
let value = value.get();
let (new_mean, new_m2) = tangram_metrics::merge_mean_m2(
current_count.to_u64().unwrap(),
mean,
m2,
count.to_u64().unwrap(),
value as f64,
0.0,
);
mean = new_mean;
m2 = new_m2;
current_count += count;
let quantiles_iter = zip!(
quantiles.iter_mut(),
quantile_indexes.iter(),
quantile_fracts.iter(),
)
.filter(|(quantile, _, _)| quantile.is_none());
for (quantile, index, fract) in quantiles_iter {
match (current_count - 1).cmp(index) {
Ordering::Equal => {
if *fract > 0.0 {
// Interpolate between two values.
let next_value = iter.peek().unwrap().0.get();
*quantile = Some(value * (1.0 - fract) + next_value * fract);
} else {
*quantile = Some(value);
}
}
Ordering::Greater => *quantile = Some(value),
Ordering::Less => {}
}
}
}
let quantiles: Vec<f32> = quantiles.into_iter().map(|q| q.unwrap()).collect();
let p25 = quantiles[0];
let p50 = quantiles[1];
let p75 = quantiles[2];
let mean = mean.to_f32().unwrap();
let variance = tangram_metrics::m2_to_variance(
m2,
NonZeroU64::new(current_count.to_u64().unwrap()).unwrap(),
);
NumberColumnStatsOutput {
column_name: self.column_name,
count: self.count,
histogram,
unique_count: unique_values_count,
max,
mean,
min,
invalid_count,
variance,
std: variance.sqrt(),
p25,
p50,
p75,
}
}
}
impl EnumColumnStats {
fn compute(
column: EnumTableColumnView,
_settings: &StatsSettings,
progress: impl Fn(u64),
) -> EnumColumnStats {
let mut histogram = vec![0; column.variants().len() + 1];
for value in column.iter() {
let index = value.map(|v| v.get()).unwrap_or(0);
histogram[index] += 1;
progress(1);
}
let invalid_count = histogram[0];
EnumColumnStats {
column_name: column.name().unwrap().to_owned(),
count: column.len(),
variants: column.variants().to_owned(),
histogram,
invalid_count,
valid_count: 0,
}
}
fn merge(mut self, other: EnumColumnStats) -> EnumColumnStats {
for (a, b) in zip!(self.histogram.iter_mut(), other.histogram.iter()) {
*a += b;
}
self.count += other.count;
self.invalid_count += other.invalid_count;
self.valid_count += other.valid_count;
self
}
fn finalize(self, _settings: &StatsSettings) -> EnumColumnStatsOutput {
EnumColumnStatsOutput {
column_name: self.column_name,
count: self.count.to_u64().unwrap(),
invalid_count: self.invalid_count,
unique_count: self.variants.len(),
histogram: zip!(self.variants, self.histogram.into_iter().skip(1))
.map(|(value, count)| (value, count))
.collect(),
}
}
}
#[derive(Clone, Debug, Eq)]
struct TokenEntry(pub NGram, pub usize);
impl std::cmp::Ord for TokenEntry {
fn cmp(&self, other: &Self) -> std::cmp::Ordering {
self.1.cmp(&other.1)
}
}
impl std::cmp::PartialOrd for TokenEntry {
fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
self.1.partial_cmp(&other.1)
}
}
impl std::cmp::PartialEq for TokenEntry {
fn eq(&self, other: &Self) -> bool {
self.1.eq(&other.1)
}
}
impl TextColumnStats {
fn compute(
column: TextTableColumnView,
settings: &StatsSettings,
progress: impl Fn(u64),
) -> TextColumnStats {
let tokenizer = Tokenizer::default();
let mut stats = TextColumnStats {
column_name: column.name().unwrap().to_owned(),
row_count: column.len(),
tokenizer,
ngrams: IndexMap::default(),
ngram_types: settings.ngram_types.to_owned(),
};
let mut ngrams_for_row = FnvHashSet::default();
for value in column.iter() {
ngrams_for_row.clear();
let unigram_iter = if stats.ngram_types.contains(&NGramType::Unigram) {
Some(
stats
.tokenizer
.tokenize(value)
.map(tangram_text::NGramRef::Unigram),
)
} else {
None
};
let bigram_iter =
if stats.ngram_types.contains(&NGramType::Bigram) {
Some(
stats.tokenizer.tokenize(value).tuple_windows().map(
|(token_a, token_b)| tangram_text::NGramRef::Bigram(token_a, token_b),
),
)
} else {
None
};
let ngram_iter = unigram_iter
.into_iter()
.flatten()
.chain(bigram_iter.into_iter().flatten());
for ngram in ngram_iter {
if let Some(entry) = stats.ngrams.get_mut(&ngram) {
entry.occurrence_count += 1;
} else {
let ngram = ngram.to_ngram();
let entry = TextColumnStatsNGramEntry {
row_count: 0,
occurrence_count: 1,
};
stats.ngrams.insert(ngram, entry);
}
ngrams_for_row.insert(ngram);
}
for ngram in ngrams_for_row.iter() {
stats.ngrams.get_mut(ngram).unwrap().row_count += 1;
}
progress(1);
}
stats
}
fn merge(mut self, other: TextColumnStats) -> TextColumnStats {
self.row_count += other.row_count;
for (other_ngram, other_entry) in other.ngrams.into_iter() {
if let Some(entry) = self.ngrams.get_mut(&other_ngram) {
entry.row_count += other_entry.row_count;
entry.occurrence_count += other_entry.occurrence_count;
} else {
self.ngrams.insert(other_ngram, other_entry);
}
}
self
}
fn finalize(mut self, settings: &StatsSettings) -> TextColumnStatsOutput {
self.ngrams
.sort_by(|_, entry_a, _, entry_b| entry_a.row_count.cmp(&entry_b.row_count));
let row_count = self.row_count.to_f32().unwrap();
let ngrams_count = self.ngrams.len();
let ngrams: IndexMap<NGram, TextColumnStatsOutputTopNGramsEntry, FnvBuildHasher> = self
.ngrams
.into_iter()
.rev()
.take(settings.ngrams_max_count)
.map(|(ngram, entry)| {
let entry_row_count = entry.row_count.to_f32().unwrap();
let idf = ((1.0 + row_count) / (1.0 + entry_row_count)).ln() + 1.0;
let entry = TextColumnStatsOutputTopNGramsEntry {
idf,
occurrence_count: entry.occurrence_count,
row_count: entry.row_count,
};
(ngram, entry)
})
.collect();
TextColumnStatsOutput {
column_name: self.column_name,
tokenizer: self.tokenizer,
row_count: self.row_count.to_u64().unwrap(),
ngram_types: settings.ngram_types.clone(),
ngrams_count,
top_ngrams: ngrams,
}
}
}