Update distillability modeling scripts to predict long articles

heuristics/distillable/check_derived_features.py

+#!/usr/bin/env python
+# Copyright 2016 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import argparse
+import csv
+import json
+import os
+import shutil
+import sys
+import unittest
+
+from write_features_csv import filter_fields, getGroups
+
+def isAlmostEqual(a, b, header, eps=0.001):
+  assert len(a) == len(b)
+  for i in range(len(a)):
+    if abs(a[i] - b[i]) > eps:
+      print '%s mismatch: a[%d] = %f, b[%d] = %f' % (header[i], i, a[i], i, b[i])
+      return False
+  return True
+
+def compareDerivedFeatures(features, from_mhtml):
+  """Compare the derived features from the JS vs. native impl
+
+  Args:
+    features: the JSON dump of features
+    from_mhtml (bool): whether the features are collected from mhtml archive
+  """
+  header = map(str, features[0]['features'][::2])
+  err = 0
+  skipped = 0
+  for f in features:
+    if not 'native' in f:
+      print 'Skipped %s' % (f['url'])
+      skipped += 1
+      continue
+    data = [map(float, f['features'][1::2])]
+    (h, data) = filter_fields(header, data, getGroups(header)['v1'])
+    js = data[0]
+    # js is now the derived features from JS aligned with native impl.
+    if not from_mhtml and js[17] != f['native']['features']['elementCount']:
+      # elementCount is simple enough so assume it's correct.
+      # If elementCount doesn't match, the DOM might've changed between JS and
+      # native runs.
+      # For mhtml, this should not be possible since DOM is static.
+      print 'Skipped %s' % (f['url'])
+      skipped += 1
+      continue
+
+    native = map(float, f['native']['derived_features'])
+    data = [js, native]
+    # Filter out the features derived from path if it is from mhtml, because
+    # the url from native impl would be the file:// one.
+    if from_mhtml:
+      (h, data) = filter_fields(h, data, getGroups(header)['v1NoPath'])
+    if not isAlmostEqual(data[0], data[1], h):
+      err += 1
+      print f['url']
+      if from_mhtml:
+        print '%s.mhtml' % f['index']
+      print data[0]
+      print data[1]
+      print
+  print '%d/%d have mismatching derived features, %d were skipped.' % (err, len(features), skipped)
+
+def main(argv):
+  parser = argparse.ArgumentParser()
+  parser.add_argument('--features', required=True, help="filename of aggregated derived features")
+  parser.add_argument('--from-mhtml', action='store_true', help="whether the features are from mhtml")
+  options = parser.parse_args(argv)
+
+  with open(options.features) as features:
+    features = json.load(features)
+    compareDerivedFeatures(features, options.from_mhtml)
+
+if __name__ == '__main__':
+  sys.exit(main(sys.argv[1:]))