import torch from transformers.models.bert.modeling_bert import BertModel, BertPreTrainedModel from torch import nn from itertools import chain from torch.nn import MSELoss, CrossEntropyLoss from cleantext import clean from num2words import num2words import re import string import inflect punct_chars = list((set(string.punctuation) | {'’', '‘', '–', '—', '~', '|', '“', '”', '…', "'", "`", '_'})) punct_chars.sort() punctuation = ''.join(punct_chars) replace = re.compile('[%s]' % re.escape(punctuation)) MATH_PREFIXES = [ "sum", "arc", "mass", "digit", "graph", "liter", "gram", "add", "angle", "scale", "data", "array", "ruler", "meter", "total", "unit", "prism", "median", "ratio", "area", # added "multipl", "divid", "subtrac", "logarit", "algebr", "calcul", "matri", "vect", "geometr", "statist", "probabli", "coeffi", "measure", "simplif" ] MATH_WORDS = [ "absolute deviation", "absolute value", "abundant number", "accurate", "acre", "acute", "add", "addend", "addition fact", "addition", "additive identity", "additive inverse", "adjacent", "algebra", "algebraic", "algorithm", "alternate interior angle", "altitude", "analog", "angle measure", "angle", "angular", "apex", "approximate", "arc", "area model", "area", "arithmetic fact", "arithmetic", "array", "associative property", "associative", "astronomical unit", "attribute", "average", "axis", "bar graph", "base of a parallelogram", "base of a prism", "base of a pyramid", "base of a triangle", "base of an exponent", "base of", "base ten", "base", "baseline", "benchmark fraction", "billion", "binomial", "bisect", "bisector", "box and whisker plot", "box plot", "capacity", "cartesian coordinate", "categorical data", "categorical", "celsius", "census", "cent", "center of a circle", "center of a dilation", "center of a sphere", "center", "centimeter", "central angle", "centroid", "chance experiment", "chance", "chord", "circle graph", "circle", "circular", "circumference", "clockwise", "coefficient", "collinear", "column matrix" "column", "combination", "combine", "common denominator", "common factor", "common fraction", "common multiple", "commutative property", "commutative", "comparison diagram", "comparison story", "compass", "complement", "complementary", "compose", "composite", "concave polygon", "concentric circles", "concentric", "cone", "congruent", "consecutive", "constant function", "constant", "continuous model of area", "continuous model of volume", "continuous", "contour", "conversion fact", "conversion factor", "convert", "convex function", "convex polygon", "coordinate", "coplanar", "corresponding", "counterclockwise", "counting numbers", "counting up subtraction", "covariance", "covariate", "cover-up method", "cross multiplication", "cross product", "cross section", "cross-section", "cube root", "cube", "cubed", "cubic unit", "cubic", "cubit", "cup", "curved surface", "customary system of measurement", "customary unit", "cylinder", "cylindrical", "data", "decagon", "decimal divisor", "decimal expanded form", "decimal fraction", "decimal point", "decimal", "decimeter", "decompose", "deficient number", "degree", "delta", "denominator", "density", "dependent event", "dependent variable", "deposit", "derivative", "determinant", "diagonal", "diameter", "difference", "differential" "digit", "digital", "dilation", "dimension", "discrete model", "displacement method", "distance", "distribution", "distributive", "divide", "divided", "divides", "dividing", "dividend", "divisibility test", "divisible by", "divisible", "division", "divisor", "dodecahedron", "dot plot", "double number line diagram", "double stem plot", "doubles fact", "edge", "egyptian multiplication", "elevation", "embed figure", "end point", "endpoint", "enlarge", "equal group", "equal part", "equal", "equality", "equation", "equidistant mark", "equilateral polygon", "equilateral triangle", "equilateral", "equivalence", "equivalent expression", "equivalent fraction", "equivalent", "error bound", "error of measurement", "estimat", "estimate", "european subtraction", "even number", "event", "expand", "expanded form", "expanded notation", "expected outcome", "expected value", "exponent", "exponential function", "exponential growth", "expression", "extended fact", "face", "fact power", "fact triangle", "factor", "factorial", "factors of number", "fahrenheit", "false number sentence", "figurate number", "flowchart", "fluid ounce", "formula", "fraction form", "fraction", "fractional part", "fractional unit", "frequency", "fulcrum", "function machine", "function", "furlong", "gallon", "gcd", "genus", "geoboard", "geometr", "geometric solid", "geometry template", "girth", "golden ratio", "golden rectangle", "gram", "graph key", "graph", "greatest common divisor" "greatest common factor", "grouping symbol", "half circle", "half-circle", "hashmark", "height of a parallelogram or triangle", "height of", "height", "hemisphere", "heptagon", "heptagonal", "hexagon", "hexagonal", "hierarchy", "histogram", "horizontal shift", "horizontal stretch", "horizontal", "hundred", "hundredth", "hypotenuse", "hypothesis", "icosahedron", "identity function", "identity matrix", "identity property of", "identity property", "improper fraction", "inch", "incircle", "indefinite integral", "independent event", "independent variable", "index of location", "indirect measurement", "inequality", "infinity", "input", "inscribed angle", "inscribed polygon", "instance of a pattern", "integer", "intercept", "intercepted arc", "interior angle", "interior of a figure", "interpolate", "interquartile range", "intersect", "interval", "inverse operation", "inverse", "iqr", "irrational number", "irrational root", "irrational", "isometry transformation", "isosceles trapezoid", "isosceles triangle", "isosceles", "joint probability", "joint variation", "juxtapose", "key sequence", "kilogram", "kilometer", "kite", "label", "landmark", "latitude", "lattice multiplication", "lcm", "least common denominator", "least common multiple", "left to right subtraction", "leg of a right triangle", "legs", "length", "like fraction", "like terms", "line graph", "line of reflection", "line of symmetry", "line plot", "line segment", "line symmetry", "line", "linear relationship", "lines of latitude", "lines of longitude", "liter", "local maximum", "local minimum", "locus", "logarithm", "logarithmic function", "logarithmic scale", "logic", "long division", "longitude", "lowest term", "magnitude estimate", "make ten", "map legend", "map scale", "mass", "maximum", "mean absolute deviation", "mean value", "mean", "measure of center", "measure", "measurement division", "measurement error", "measurement unit", "median", "meridian bar", "meter", "meters per second", "metric system", "metric unit", "metric", "midpoint", "mile", "milliliter", "millimeter", "millisecond", "minimum", "minuend", "mirror image", "mixed number", "mixed unit", "mobius", "modal", "mode", "multipl", "multiply", "multiplied", "multiplies", "multiple", "multiplication", "multiplying", "multiplication counting principle", "multiplication diagram", "multiplication fact", "multiplication symbol", "multiplication use class", "multiplicative identity", "multiplicative inverse", "multiplier", "mutually exclusive event", "natural number", "negative association", "negative exponent", "negative number", "negative rational number", "nested parentheses", "net score", "net weight", "net", "nonagon", "nonconvex polygon", "nonlinear", "normal distribution", "normal span", "normal", "number bond", "number disk", "number grid", "number line", "number path", "number sentence", "number sequence", "numeral", "numeration", "numerator", "numerical data", "numerical", "obtuse", "octagon", "octagonal", "octahedron", "odd number", "open proportion", "operation symbol", "operational", "opposite angle", "opposite change rule", "opposite of a number", "opposite side", "opposite vertex", "opposite", "order of magnitude", "order of operations", "order of rotation symmetry", "order of", "ordered pair", "ordered", "ordinal number", "orthogonal", "ounce", "outlier", "pace", "pan balance", "parabola", "parallel lines", "parallel plane", "parallel", "parallelogram", "parentheses", "part to part ratio", "part to whole ratio", "part whole fraction", "partial differences subtraction", "partial product", "partial products multiplication", "partial quotients division", "partial sums addition", "partition", "partitive division", "parts and total diagram", "pentagon", "pentagonal", "per capita", "per unit rate", "per", "percent circle", "percent", "percentage", "perfect number", "perfect square", "perfect triangle", "perimeter", "permutation", "perpendicular", "perpetual calendar", "pi", "picture graph", "pie graph", "pint", "pivot", "place value", "plane figure", "plane", "point symmetry", "point", "polar coordinate", "polygon", "polyhedron", "polynominal" "population density", "population", "positive association", "positive number", "pound", "power", "precise", "predict", "prediction line", "preimage", "prime factor", "prime factorization", "prime meridian", "prime number", "prism", "probability meter", "probability tree diagram", "probability", "product", "proper factor", "proper fraction", "property", "proportion", "proportional", "proportionality", "protractor", "pyramid", "pythagorean theorem", "quadrangle", "quadrant", "quadratic", "quadrilateral", "quart", "quarter circle", "quarter of", "quarter-circle", "quartile", "quick common denominator", "quotient", "quotitive division", "radian", "radius of" "radius", "random draw", "random experiment", "random number", "random sample", "random", "range", "rank", "rate diagram", "rate multiplication ", "rate of change", "rate unit", "rate", "ratio of", "ratio", "rational equation", "rational number", "ray", "real number", "recall survey", "reciprocal", "rectang", "rectangle", "rectangular array", "rectangular coordinate grid", "rectangular prism", "rectangular pyramid", "rectangular", "rectilinear figure", "reflection", "reflex angle", "region", "regular polygon", "regular polyhedron", "regular tessellation", "relation symbol", "relative frequency", "remainder", "repeated addition", "repeating decimal", "representative", "revolution", "rhombus", "right angle", "right cone", "right cylinder", "right prism", "right pyramid", "right triangle", "rigid transformation", "roman numerals", "root", "rotate", "rotation symmetry", "rotation", "round off", "round-off", "ruler", "same change rule for subtraction", "sample", "scalar", "scale factor", "scale model", "scale of a map", "scale of a number line", "scale", "scaled graph", "scaled", "scalene triangle", "scalene", "scatter plot", "scattergram", "sector", "segment", "semi-circle", "semicircle", "sequence", "set", "sign", "significant digit", "significant figure", "similar figures", "similar", "simpler form", "simplify", "simulation", "situtation diagram", "skew line", "slanted", "slide rule", "slope", "solid figure", "solution", "span", "speed", "sphere", "square root", "square unit", "square", "squared", "stacked bar graph", "standard form", "standard unit", "statistic", "stem and leaf plot", "step graph", "straight angle", "straightedge", "subset of" "substitute", "subtract", "subtrahend", "sum of", "sum", "supplementary angle", "surface area", "surface", "survey", "symmetric", "symmetry", "system of equation", "system of", "table", "take from ten", "tally", "tangent circle", "tangent", "tangram", "tape diagram", "temperature", "template", "tens place", "tenth", "term", "terminating decimal", "tessellat", "tessellate", "tessellation", "tetrahedron", "tetromino", "theorem", "thermometer", "thousand", "thousandth", "tile", "tiling", "time graph", "timeline", "top heavy fraction", "topological", "topology", "total area", "total of", "total surface", "total volume", "trade first subtraction", "transformation", "translation", "transversal", "trapezoid", "tree diagram", "triangle", "triangular", "true number sentence", "truncate", "twin prime", "two-way table", "unit cube", "unit form", "unit fraction", "unit interval", "unit price", "unit rate", "unit square", "unit", "unknown", "unlike denominator", "unlike fraction", "value", "vanishing ", "variability", "variable", "velocity", "venn diagram", "vernal equinox", "vertex", "vertical", "volume of", "volume", "weight", "whole number", "whole unit", "whole", "width", "withdrawal", "word form", "x axes", "x axis", "x intercept", "x-axes", "x-axis", "y axes", "y axis", "y intercept", "y-axes", "y-axis", "y-intercept", "yard", "zero property of multiplication", "zero", ] PLURAL_TO_SINGULAR_EXCLUSIONS = [ "data", "minus", "plus", ] p = inflect.engine() def singular_to_plural(word): """Convert singular words to plural using inflect.""" plural = p.plural(word) return plural or word def plural_to_singular(word): """Convert plural word to singular using inflect.""" if word in PLURAL_TO_SINGULAR_EXCLUSIONS: return word return p.singular_noun(word) or word plural_MATH_WORDS = [singular_to_plural(word) for word in MATH_WORDS] MATH_WORDS += plural_MATH_WORDS def get_num_words(text): if not isinstance(text, str): print("%s is not a string" % text) text = replace.sub(' ', text) text = re.sub(r'\s+', ' ', text) text = text.strip() text = re.sub(r'\[.+\]', " ", text) return len(text.split()) def number_to_words(num): try: return num2words(re.sub(",", "", num)) except: return num clean_str = lambda s: clean(s, fix_unicode=True, # fix various unicode errors to_ascii=True, # transliterate to closest ASCII representation lower=True, # lowercase text no_line_breaks=True, # fully strip line breaks as opposed to only normalizing them no_urls=True, # replace all URLs with a special token no_emails=True, # replace all email addresses with a special token no_phone_numbers=True, # replace all phone numbers with a special token no_numbers=True, # replace all numbers with a special token no_digits=False, # replace all digits with a special token no_currency_symbols=False, # replace all currency symbols with a special token no_punct=False, # fully remove punctuation replace_with_url="", replace_with_email="", replace_with_phone_number="", replace_with_number=lambda m: number_to_words(m.group()), replace_with_digit="0", replace_with_currency_symbol="", lang="en" ) clean_str_nopunct = lambda s: clean(s, fix_unicode=True, # fix various unicode errors to_ascii=True, # transliterate to closest ASCII representation lower=True, # lowercase text no_line_breaks=True, # fully strip line breaks as opposed to only normalizing them no_urls=True, # replace all URLs with a special token no_emails=True, # replace all email addresses with a special token no_phone_numbers=True, # replace all phone numbers with a special token no_numbers=True, # replace all numbers with a special token no_digits=False, # replace all digits with a special token no_currency_symbols=False, # replace all currency symbols with a special token no_punct=True, # fully remove punctuation replace_with_url="", replace_with_email="", replace_with_phone_number="", replace_with_number=lambda m: number_to_words(m.group()), replace_with_digit="0", replace_with_currency_symbol="", lang="en" ) class MultiHeadModel(BertPreTrainedModel): """Pre-trained BERT model that uses our loss functions""" def __init__(self, config, head2size): super(MultiHeadModel, self).__init__(config, head2size) config.num_labels = 1 self.bert = BertModel(config) self.dropout = nn.Dropout(config.hidden_dropout_prob) module_dict = {} for head_name, num_labels in head2size.items(): module_dict[head_name] = nn.Linear(config.hidden_size, num_labels) self.heads = nn.ModuleDict(module_dict) self.init_weights() def forward(self, input_ids, token_type_ids=None, attention_mask=None, head2labels=None, return_pooler_output=False, head2mask=None, nsp_loss_weights=None): device = "cuda" if torch.cuda.is_available() else "cpu" # Get logits output = self.bert( input_ids, token_type_ids=token_type_ids, attention_mask=attention_mask, output_attentions=False, output_hidden_states=False, return_dict=True) pooled_output = self.dropout(output["pooler_output"]).to(device) head2logits = {} return_dict = {} for head_name, head in self.heads.items(): head2logits[head_name] = self.heads[head_name](pooled_output) head2logits[head_name] = head2logits[head_name].float() return_dict[head_name + "_logits"] = head2logits[head_name] if head2labels is not None: for head_name, labels in head2labels.items(): num_classes = head2logits[head_name].shape[1] # Regression (e.g. for politeness) if num_classes == 1: # Only consider positive examples if head2mask is not None and head_name in head2mask: num_positives = head2labels[head2mask[head_name]].sum() # use certain labels as mask if num_positives == 0: return_dict[head_name + "_loss"] = torch.tensor([0]).to(device) else: loss_fct = MSELoss(reduction='none') loss = loss_fct(head2logits[head_name].view(-1), labels.float().view(-1)) return_dict[head_name + "_loss"] = loss.dot(head2labels[head2mask[head_name]].float().view(-1)) / num_positives else: loss_fct = MSELoss() return_dict[head_name + "_loss"] = loss_fct(head2logits[head_name].view(-1), labels.float().view(-1)) else: loss_fct = CrossEntropyLoss(weight=nsp_loss_weights.float()) return_dict[head_name + "_loss"] = loss_fct(head2logits[head_name], labels.view(-1)) if return_pooler_output: return_dict["pooler_output"] = output["pooler_output"] return return_dict class InputBuilder(object): """Base class for building inputs from segments.""" def __init__(self, tokenizer): self.tokenizer = tokenizer self.mask = [tokenizer.mask_token_id] def build_inputs(self, history, reply, max_length): raise NotImplementedError def mask_seq(self, sequence, seq_id): sequence[seq_id] = self.mask return sequence @classmethod def _combine_sequence(self, history, reply, max_length, flipped=False): # Trim all inputs to max_length history = [s[:max_length] for s in history] reply = reply[:max_length] if flipped: return [reply] + history return history + [reply] class BertInputBuilder(InputBuilder): """Processor for BERT inputs""" def __init__(self, tokenizer): InputBuilder.__init__(self, tokenizer) self.cls = [tokenizer.cls_token_id] self.sep = [tokenizer.sep_token_id] self.model_inputs = ["input_ids", "token_type_ids", "attention_mask"] self.padded_inputs = ["input_ids", "token_type_ids"] self.flipped = False def build_inputs(self, history, reply, max_length, input_str=True): """See base class.""" if input_str: history = [self.tokenizer.convert_tokens_to_ids(self.tokenizer.tokenize(t)) for t in history] reply = self.tokenizer.convert_tokens_to_ids(self.tokenizer.tokenize(reply)) sequence = self._combine_sequence(history, reply, max_length, self.flipped) sequence = [s + self.sep for s in sequence] sequence[0] = self.cls + sequence[0] instance = {} instance["input_ids"] = list(chain(*sequence)) last_speaker = 0 other_speaker = 1 seq_length = len(sequence) instance["token_type_ids"] = [last_speaker if ((seq_length - i) % 2 == 1) else other_speaker for i, s in enumerate(sequence) for _ in s] return instance