جواب «Step 4: The Best Answer» چالش 4 از جلسهی 58
سؤال
قدم چهارم: تابع best_match(question) را کامل کن که شمارهی درسی با بیشترین امتیاز و خود امتیاز را برگرداند. answer (آماده) جواب آمادهی همان درس را میدهد. خروجی چهار خط: Q: What does softmax do with scores?، Bot: Softmax turns scores into probabilities that add up to 1.، Q: Why do I need a validation set?، Bot: A validation set is kept apart from training so you can measure the model fairly.
راهنما
روی range(len(s)) بچرخ و اگر s[i] > s[best] بود، best = i. یا int(np.argmax(s)).
یک جواب درست
import numpy as np
LESSONS = [
{"title": "overfitting", "source": "session 48",
"text": "overfitting is when a model memorises the training data and does badly on new test data",
"answer": "Overfitting means the model memorised its training examples and fails on new ones."},
{"title": "validation", "source": "session 49",
"text": "a validation set is data the model never trains on so we can check it fairly",
"answer": "A validation set is kept apart from training so you can measure the model fairly."},
{"title": "relu", "source": "session 43",
"text": "relu is an activation that keeps positive numbers and turns negative numbers into zero",
"answer": "ReLU keeps a positive number and turns a negative one into 0, like max(0, x)."},
{"title": "softmax", "source": "session 44",
"text": "softmax turns a list of scores into probabilities that are positive and add up to one",
"answer": "Softmax turns scores into probabilities that add up to 1."},
{"title": "mini-batch", "source": "session 45",
"text": "a mini batch is a small group of training examples used for one step of gradient descent",
"answer": "A mini-batch is a small group of examples used for one gradient descent step."},
{"title": "decision tree", "source": "sessions 50-51",
"text": "a decision tree asks yes or no questions about the features and splits the data at every node",
"answer": "A decision tree asks one yes/no question per node and splits the data until a leaf answers."},
{"title": "convolution", "source": "session 53",
"text": "convolution slides a small filter over the image and adds up each patch times the filter to find edges",
"answer": "Convolution slides a small filter across the picture and adds up patch * filter at every spot."},
{"title": "embedding", "source": "session 55",
"text": "a word embedding is a vector of numbers for a word so that similar words get close vectors",
"answer": "A word embedding is a vector for a word; similar words point in similar directions."},
{"title": "attention", "source": "session 57",
"text": "attention lets every word look at the other words of the sentence with a query a key and a value",
"answer": "In attention every word scores the others (query times key), softmax gives weights, and it mixes their values."},
]
def show_lessons():
print(len(LESSONS), "lessons")
for lesson in LESSONS:
print(f"{lesson['title']} ({lesson['source']})")
STOP = {"a", "an", "the", "is", "are", "what", "how", "does", "do", "of", "to", "and", "in", "on",
"it", "that", "for", "with", "at", "so", "by", "i", "my", "we", "can", "you", "why", "which",
"into", "its", "be", "this", "about", "or", "when", "me", "every", "each"}
def tokenize(text):
clean = "".join(ch if ch.isalpha() else " " for ch in text.lower())
return [w for w in clean.split() if w not in STOP]
VOCAB = sorted(set(w for lesson in LESSONS for w in tokenize(lesson["text"])))
def vectorize(text):
v = np.zeros(len(VOCAB))
for w in tokenize(text):
if w in VOCAB:
v[VOCAB.index(w)] += 1
return v
LESSON_VECTORS = [vectorize(lesson["text"]) for lesson in LESSONS]
def cosine(a, b):
na = np.linalg.norm(a)
nb = np.linalg.norm(b)
if na == 0 or nb == 0:
return 0.0
return float(a @ b / (na * nb))
def scores(question):
q = vectorize(question)
return [cosine(q, v) for v in LESSON_VECTORS]
def best_match(question):
s = scores(question)
best = 0
for i in range(len(s)):
if s[i] > s[best]:
best = i
return best, s[best]
def answer(question):
i, score = best_match(question)
return LESSONS[i]["answer"]
for q in ["What does softmax do with scores?", "Why do I need a validation set?"]:
print("Q:", q)
print("Bot:", answer(q))
اول خودت امتحان کن؛ جوابهای دیگر هم قبول است. برگرد و خودت حلش کن