What is a Dictionary?

What you need to know

The basic operations

1user = {"name": "Asha", "role": "ML Engineer"}23print(user["role"])              # ML Engineer4print(user.get("city", "NA"))    # NA  -> no KeyError for a missing key5user["city"] = "Pune"            # insert or update6del user["role"]                 # delete7print("city" in user)            # True  -> checks KEYS, not values8print(user)                      # {'name': 'Asha', 'city': 'Pune'}9print(user | {"team": "search"}) # {'name': 'Asha', 'city': 'Pune', 'team': 'search'}

The | merge operator creates a new dict (Python 3.9+). Writing to an existing key replaces the value, which is why keys are unique.

How a hash table gives O(1)

When you write user["city"], Python computes hash("city"), uses it to pick a slot in an internal array, checks that the key stored there really equals "city", and returns the value. It does not scan the other keys. With a list of pairs you would compare key after key — O(n). With a dict the time stays about the same for 10 keys or 10 million.

Two keys can land in the same slot (a collision); Python then probes other slots. Collisions are rare with a good hash, so the average stays O(1), though the theoretical worst case is O(n).

Why keys must be hashable

If a key could change after insertion, its hash would change and Python would look in the wrong slot. So keys must be immutable types: str, int, float, tuple of hashables, frozenset. A list or dict cannot be a key.

Ordering

Since Python 3.7 the language guarantees that iteration follows insertion order. Older answers saying "dicts are unordered" are out of date.

A real-life example

You have a day's transactions and want the count and total per merchant — the core of any spending dashboard:

1txns = [("Swiggy", 349), ("Zomato", 512), ("Swiggy", 180), ("IRCTC", 1450), ("Swiggy", 99)]23stats = {}4for merchant, amount in txns:5    entry = stats.get(merchant, {"count": 0, "total": 0})6    entry["count"] += 17    entry["total"] += amount8    stats[merchant] = entry910for merchant, s in stats.items():11    print(merchant, s)12# Swiggy {'count': 3, 'total': 628}13# Zomato {'count': 1, 'total': 512}14# IRCTC {'count': 1, 'total': 1450}

Each row costs one O(1) lookup, so 5 rows or 5 million rows scale linearly. The same shape powers LLM calls: messages = [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}] is a list of dicts.

Follow-up questions to expect

• "What happens on a hash collision?" — Python probes for another free slot and compares keys with == to find the right one. Lookups stay O(1) on average.
• "dict vs defaultdict vs Counter?" — defaultdict(list) creates a default value for missing keys automatically; Counter is a dict specialised for counting, with helpers like most_common(3).
• "Can two keys be 1 and 1.0?" — No. 1 == 1.0 and they hash the same, so they are the same key.