Sign upLogin

tensorflow/models

View on GitHub
Star
official/legacy/transformer/utils/metrics.py

Summary

Maintainability
F
1 wk
Test Coverage

File metrics.py has 365 lines of code (exceeds 250 allowed). Consider refactoring.
Open

# Copyright 2024 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
Severity: Minor
Found in official/legacy/transformer/utils/metrics.py - About 4 hrs to fix

    Function compute_bleu has a Cognitive Complexity of 17 (exceeds 5 allowed). Consider refactoring.
    Open

    def compute_bleu(reference_corpus, translation_corpus, max_order=4,
                 use_bp=True):
  """Computes BLEU score of translated segments against one or more references.

  Args:
    Severity: Minor
    Found in official/legacy/transformer/utils/metrics.py - About 2 hrs to fix

    Cognitive Complexity

    Cognitive Complexity is a measure of how difficult a unit of code is to intuitively understand. Unlike Cyclomatic Complexity, which determines how difficult your code will be to test, Cognitive Complexity tells you how difficult your code will be to read and comprehend.

    A method's cognitive complexity is based on a few simple rules:

    • Code is not considered more complex when it uses shorthand that the language provides for collapsing multiple statements into one
    • Code is considered more complex for each "break in the linear flow of the code"
    • Code is considered more complex when "flow breaking structures are nested"

    Further reading

    Function _lcs has a Cognitive Complexity of 9 (exceeds 5 allowed). Consider refactoring.
    Open

    def _lcs(x, y):
  """Computes the length of the LCS between two seqs.

  The implementation below uses a DP programming algorithm and runs
  in O(nm) time where n = len(x) and m = len(y).
    Severity: Minor
    Found in official/legacy/transformer/utils/metrics.py - About 55 mins to fix

    Cognitive Complexity

    Cognitive Complexity is a measure of how difficult a unit of code is to intuitively understand. Unlike Cyclomatic Complexity, which determines how difficult your code will be to test, Cognitive Complexity tells you how difficult your code will be to read and comprehend.

    A method's cognitive complexity is based on a few simple rules:

    • Code is not considered more complex when it uses shorthand that the language provides for collapsing multiple statements into one
    • Code is considered more complex for each "break in the linear flow of the code"
    • Code is considered more complex when "flow breaking structures are nested"

    Further reading

    Function rouge_n has a Cognitive Complexity of 7 (exceeds 5 allowed). Consider refactoring.
    Open

    def rouge_n(eval_sentences, ref_sentences, n=2):
  """Computes ROUGE-N f1 score of two text collections of sentences.

  Source: https://www.microsoft.com/en-us/research/publication/
  rouge-a-package-for-automatic-evaluation-of-summaries/
    Severity: Minor
    Found in official/legacy/transformer/utils/metrics.py - About 35 mins to fix

    Cognitive Complexity

    Cognitive Complexity is a measure of how difficult a unit of code is to intuitively understand. Unlike Cyclomatic Complexity, which determines how difficult your code will be to test, Cognitive Complexity tells you how difficult your code will be to read and comprehend.

    A method's cognitive complexity is based on a few simple rules:

    • Code is not considered more complex when it uses shorthand that the language provides for collapsing multiple statements into one
    • Code is considered more complex for each "break in the linear flow of the code"
    • Code is considered more complex when "flow breaking structures are nested"

    Further reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

      with tf.variable_scope("padded_accuracy_topk", values=[logits, labels]):
    logits, labels = _pad_tensors_to_same_length(logits, labels)
    weights = tf.cast(tf.not_equal(labels, 0), tf.float32)
    effective_k = tf.minimum(k, tf.shape(logits)[-1])
    _, outputs = tf.nn.top_k(logits, k=effective_k)
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 1 day to fix
    official/legacy/transformer/metrics.py on lines 96..107

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 177.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

      for (references, translations) in zip(reference_corpus, translation_corpus):
    reference_length += len(references)
    translation_length += len(translations)
    ref_ngram_counts = _get_ngrams_with_counter(references, max_order)
    translation_ngram_counts = _get_ngrams_with_counter(translations, max_order)
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 1 day to fix
    official/nlp/metrics/bleu.py on lines 132..145

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 164.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

      with tf.variable_scope("padded_sequence_accuracy", values=[logits, labels]):
    logits, labels = _pad_tensors_to_same_length(logits, labels)
    weights = tf.cast(tf.not_equal(labels, 0), tf.float32)
    outputs = tf.cast(tf.argmax(logits, axis=-1), tf.int32)
    padded_labels = tf.cast(labels, tf.int32)
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 1 day to fix
    official/legacy/transformer/metrics.py on lines 116..125

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 162.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Similar blocks of code found in 2 locations. Consider refactoring.
    Open

      for i in xrange(0, max_order):
    if possible_matches_by_order[i] > 0:
      precisions[i] = float(matches_by_order[i]) / possible_matches_by_order[i]
      if matches_by_order[i] > 0:
        precisions[i] = float(matches_by_order[i]) / possible_matches_by_order[
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 1 day to fix
    official/nlp/metrics/bleu.py on lines 150..160

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 158.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

    def _pad_tensors_to_same_length(x, y):
  """Pad x and y so that the results have the same length (second dimension)."""
  with tf.name_scope("pad_to_same_length"):
    x_length = tf.shape(x)[1]
    y_length = tf.shape(y)[1]
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 1 day to fix
    official/legacy/transformer/metrics.py on lines 31..41

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 130.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Similar blocks of code found in 2 locations. Consider refactoring.
    Open

    def _get_ngrams_with_counter(segment, max_order):
  """Extracts all n-grams up to a given maximum order from an input segment.

  Args:
    segment: text segment from which n-grams will be extracted.
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 6 hrs to fix
    official/nlp/metrics/bleu.py on lines 87..103

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 102.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

      with tf.variable_scope("padded_accuracy", values=[logits, labels]):
    logits, labels = _pad_tensors_to_same_length(logits, labels)
    weights = tf.cast(tf.not_equal(labels, 0), tf.float32)
    outputs = tf.cast(tf.argmax(logits, axis=-1), tf.int32)
    padded_labels = tf.cast(labels, tf.int32)
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 5 hrs to fix
    official/legacy/transformer/metrics.py on lines 86..91

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 95.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Similar blocks of code found in 3 locations. Consider refactoring.
    Open

    def rouge_2_fscore(logits, labels):
  """ROUGE-2 F1 score computation between labels and predictions.

  This is an approximate ROUGE scoring method since we do not glue word pieces
  or decode the ids and tokenize the output.
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 2 other locations - About 3 hrs to fix
    official/legacy/transformer/utils/metrics.py on lines 191..208
    official/legacy/transformer/utils/metrics.py on lines 374..390

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 64.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Similar blocks of code found in 3 locations. Consider refactoring.
    Open

    def bleu_score(logits, labels):
  """Approximate BLEU score computation between labels and predictions.

  An approximate BLEU scoring method since we do not glue word pieces or
  decode the ids and tokenize the output. By default, we use ngram order of 4
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 2 other locations - About 3 hrs to fix
    official/legacy/transformer/utils/metrics.py on lines 297..313
    official/legacy/transformer/utils/metrics.py on lines 374..390

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 64.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Similar blocks of code found in 3 locations. Consider refactoring.
    Open

    def rouge_l_fscore(predictions, labels):
  """ROUGE scores computation between labels and predictions.

  This is an approximate ROUGE scoring method since we do not glue word pieces
  or decode the ids and tokenize the output.
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 2 other locations - About 3 hrs to fix
    official/legacy/transformer/utils/metrics.py on lines 191..208
    official/legacy/transformer/utils/metrics.py on lines 297..313

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 64.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

      if max(precisions) > 0:
    p_log_sum = sum(math.log(p) for p in precisions if p)
    geo_mean = math.exp(p_log_sum / max_order)
    Severity: Major
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 2 hrs to fix
    official/nlp/metrics/bleu.py on lines 162..164

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 56.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

    def padded_neg_log_perplexity(logits, labels, vocab_size):
  """Average log-perplexity excluding padding 0s. No smoothing."""
  num, den = padded_cross_entropy_loss(logits, labels, 0, vocab_size)
  return -num, den
    Severity: Minor
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 50 mins to fix
    official/legacy/transformer/metrics.py on lines 128..131

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 36.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

          low_confidence = (1.0 - confidence) / tf.cast(vocab_size - 1, tf.float32)
    Severity: Minor
    Found in official/legacy/transformer/utils/metrics.py and 1 other location - About 30 mins to fix
    official/legacy/transformer/metrics.py on lines 63..63

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 32.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    There are no issues that match your filters.

    Category
    Status