jantrienes

Tranformers Text Classifier Example

Was used to train this classifier: https://huggingface.co/jantrienes/roberta-large-question-classifier

jantrienes

import itertools
from pprint import pprint
from typing import List, Tuple


def jaccard_similarity(list1: List[str], list2: List[str]):
    if not list1 or not list2:
        return 0
    s1 = set(list1)
    s2 = set(list2)

jantrienes

from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns

def plot_confusion_matrix(y_true, y_pred, class_names, normalize=None,
                          title='Confusion Matrix', plot_numbers=False, display_names=None,
                          figsize=(15, 11)):

    cm = confusion_matrix(y_true, y_pred, labels=class_names, normalize=normalize)

jantrienes

%matplotlib inline
%config InlineBackend.figure_format='retina'

import matplotlib.pyplot as plt
import matplotlib.ticker as mtick
import seaborn as sns
sns.set_color_codes()
sns.set_theme()
sns.set_context("paper")

jantrienes

import pandas as pd
from sklearn.compose import ColumnTransformer
from sklearn.datasets import fetch_20newsgroups
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import MinMaxScaler

import eli5

jantrienes

from functools import wraps
from timeit import default_timer as timer

def timing(f):
    @wraps(f)
    def wrap(*args, **kw):
        ts = timer()
        result = f(*args, **kw)
        te = timer()
        print('func: {} took: {:2.4f} sec'.format(f.__name__, te - ts))

jantrienes

Visualize Precision/Recall Tradeoff at Operating Levels

Sometimes, it can be interesting to visualize precision vs. recall at different operating levels. For example, in a binary classification problem one can adjust the default classification threshold of the positive class (i.e., T = 0.5) to be either more or less conservative. This in turn changes precision and recall.

This utility allows to visualize the precision/recall that could be achieved when setting the classification threshold to a desired level of precision/recall.

Example:

from matplotlib import rcParams

jantrienes

def interleave(list_a, list_b):
    return _interleave(list_a, list_b, 0, 0, list())

def _interleave(list_a, list_b, k_a, k_b, combined):
    if k_a >= len(list_a) and k_b >= len(list_b):
        return combined

    if k_a == k_b:
        if list_a[k_a] not in combined:
            combined.append(list_a[k_a])

jantrienes

\documentclass[tikz]{standalone}

\begin{document}

\begin{tikzpicture}
[   cnode/.style={draw=black,fill=#1,minimum width=3mm,circle},
]
	% output neuron    
    \node[cnode=white, label=above:$\delta_1^{(2)}$, label=below:$y_1$] (s) at (6,-2) {};

jantrienes

import React from 'react';
import {
    gql,
    graphql,
} from 'react-apollo';

import AddChannel from './AddChannel';

const ChannelsList = ({ data: {loading, error, channels }}) => {
  if (loading) {