234 lines
		
	
	
	
		
			6.1 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
			
		
		
	
	
			234 lines
		
	
	
	
		
			6.1 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
# After stumbling through my first technical screen, I'm going to drill
 | 
						|
# algorithms for implementing evaluators for a toy expression language:
 | 
						|
# e.g. 2 + 13 * 3 + 5 * 2
 | 
						|
#
 | 
						|
# As of now, I'm aware of a few algorithms for solving this:
 | 
						|
#   - DONE: Convert infix expression to Polish notation and evaluate the Polish
 | 
						|
#     notation.
 | 
						|
#   - DONE: Evaluate the tokens using two stacks and avoid converting it.
 | 
						|
#   - DONE: Create a tree of depth two to encode the operator precedence and
 | 
						|
#     evaluate that AST.
 | 
						|
#   - TODO: Convert the infix expression to a prefix expression
 | 
						|
#   - TODO: Write a recursive descent parser and evaluate the AST.
 | 
						|
 | 
						|
operators = {
 | 
						|
    '*': 1,
 | 
						|
    '+': 0,
 | 
						|
}
 | 
						|
 | 
						|
def tokenize(xs):
 | 
						|
    result = []
 | 
						|
    i = 0
 | 
						|
    while i < len(xs):
 | 
						|
        current = xs[i]
 | 
						|
        if current == ' ':
 | 
						|
            i += 1
 | 
						|
            continue
 | 
						|
        elif current in operators.keys():
 | 
						|
            result.append(current)
 | 
						|
            i += 1
 | 
						|
        else:
 | 
						|
            i += 1
 | 
						|
            while i < len(xs) and xs[i] in {str(n) for n in range(10)}:
 | 
						|
                current += xs[i]
 | 
						|
                i += 1
 | 
						|
            result.append(int(current))
 | 
						|
    return result
 | 
						|
 | 
						|
# Convert infix to postfix; evaluate postfix
 | 
						|
# I believe this is known as the Shunting-Yards algorithm
 | 
						|
def postfix(tokens):
 | 
						|
    result = []
 | 
						|
    s = []
 | 
						|
    for token in tokens:
 | 
						|
        if type(token) == int:
 | 
						|
            result.append(token)
 | 
						|
        else:
 | 
						|
            while s and operators[token] < operators[s[-1]]:
 | 
						|
                result.append(s.pop())
 | 
						|
            s.append(token)
 | 
						|
    while s:
 | 
						|
        result.append(s.pop())
 | 
						|
    return result
 | 
						|
 | 
						|
def do_evaluate_with_polish_notation(tokens):
 | 
						|
    s = []
 | 
						|
    for token in tokens:
 | 
						|
        if token == '*':
 | 
						|
            s.append(s.pop() * s.pop())
 | 
						|
        elif token == '+':
 | 
						|
            s.append(s.pop() + s.pop())
 | 
						|
        else:
 | 
						|
            s.append(token)
 | 
						|
    return s[-1]
 | 
						|
 | 
						|
def evaluate_with_polish_notation(expr):
 | 
						|
    tokens = tokenize(expr)
 | 
						|
    print("Tokens:  {}".format(tokens))
 | 
						|
    pn = postfix(tokens)
 | 
						|
    print("Postfix: {}".format(pn))
 | 
						|
    result = do_evaluate_with_polish_notation(pn)
 | 
						|
    print("Result:  {}".format(result))
 | 
						|
    return result
 | 
						|
 | 
						|
# Evaluate Tokens
 | 
						|
 | 
						|
def apply_operator(op, a, b):
 | 
						|
    if op == '*':
 | 
						|
        return a * b
 | 
						|
    elif op == '+':
 | 
						|
        return a + b
 | 
						|
 | 
						|
def do_evaluate_tokens(tokens):
 | 
						|
    vals = []
 | 
						|
    ops = []
 | 
						|
    for token in tokens:
 | 
						|
        if type(token) == int:
 | 
						|
            vals.append(token)
 | 
						|
        elif token == '*':
 | 
						|
            ops.append(token)
 | 
						|
        elif token == '+':
 | 
						|
            while ops and operators[token] < operators[ops[-1]]:
 | 
						|
                vals.append(apply_operator(ops.pop(), vals.pop(), vals.pop()))
 | 
						|
            ops.append(token)
 | 
						|
        else:
 | 
						|
            raise Exception("Unexpected token: {}".format(token))
 | 
						|
    while ops:
 | 
						|
        vals.append(apply_operator(ops.pop(), vals.pop(), vals.pop()))
 | 
						|
    return vals[-1]
 | 
						|
 | 
						|
def evaluate_tokens(expr):
 | 
						|
    tokens = tokenize(expr)
 | 
						|
    print("Tokens:  {}".format(tokens))
 | 
						|
    result = do_evaluate_tokens(tokens)
 | 
						|
    print("Result:  {}".format(result))
 | 
						|
    return result
 | 
						|
 | 
						|
# Ad Hoc Tree
 | 
						|
 | 
						|
def parse(tokens):
 | 
						|
    result = []
 | 
						|
    series = []
 | 
						|
    for token in tokens:
 | 
						|
        if type(token) == int:
 | 
						|
            series.append(token)
 | 
						|
        elif token == '*':
 | 
						|
            continue
 | 
						|
        elif token == '+':
 | 
						|
            result.append(series)
 | 
						|
            series = []
 | 
						|
        else:
 | 
						|
            raise Exception("Unexpected token: {}".format(token))
 | 
						|
    result.append(series)
 | 
						|
    return result
 | 
						|
 | 
						|
def product(xs):
 | 
						|
    result = 1
 | 
						|
    for x in xs:
 | 
						|
        result *= x
 | 
						|
    return result
 | 
						|
 | 
						|
def do_evaluate_ad_hoc_tree(ast):
 | 
						|
    return sum([product(xs) for xs in ast])
 | 
						|
 | 
						|
def evaluate_ad_hoc_tree(expr):
 | 
						|
    tokens = tokenize(expr)
 | 
						|
    print("Tokens:  {}".format(tokens))
 | 
						|
    ast = parse(tokens)
 | 
						|
    print("AST:     {}".format(ast))
 | 
						|
    result = do_evaluate_ad_hoc_tree(ast)
 | 
						|
    print("Result:  {}".format(result))
 | 
						|
    return result
 | 
						|
 | 
						|
# Recursive Descent Parser
 | 
						|
 | 
						|
# expression     -> addition ;
 | 
						|
# addition       -> multiplication ( "+" multiplication )* ;
 | 
						|
# multiplication -> terminal ( "*" terminal )* ;
 | 
						|
# terminal       -> NUMBER ;
 | 
						|
 | 
						|
class Parser(object):
 | 
						|
    def __init__(self, tokens):
 | 
						|
        self.tokens = tokens
 | 
						|
        self.i = 0
 | 
						|
 | 
						|
    # mutations
 | 
						|
    def advance(self):
 | 
						|
        self.i += 1
 | 
						|
 | 
						|
    def consume(self):
 | 
						|
        result = self.curr()
 | 
						|
        self.advance()
 | 
						|
        return result
 | 
						|
 | 
						|
    # predicates
 | 
						|
    def match(self, x):
 | 
						|
        if self.curr() == x:
 | 
						|
            self.advance()
 | 
						|
            return True
 | 
						|
        return False
 | 
						|
 | 
						|
    def tokens_available(self):
 | 
						|
        return self.i < len(self.tokens)
 | 
						|
 | 
						|
    # getters
 | 
						|
    def prev(self):
 | 
						|
        return self.tokens[self.i - 1]
 | 
						|
 | 
						|
    def curr(self):
 | 
						|
        return self.tokens[self.i] if self.tokens_available() else None
 | 
						|
 | 
						|
    def next(self):
 | 
						|
        return self.tokens[self.i + 1]
 | 
						|
 | 
						|
def parse_expression(tokens):
 | 
						|
    parser = Parser(tokens)
 | 
						|
    return parse_addition(parser)
 | 
						|
 | 
						|
def parse_addition(parser):
 | 
						|
    result = parse_multiplication(parser)
 | 
						|
    while parser.match("+"):
 | 
						|
        op = parser.prev()
 | 
						|
        rhs = parse_multiplication(parser)
 | 
						|
        result = ["+", result, rhs]
 | 
						|
    return result
 | 
						|
 | 
						|
def parse_multiplication(parser):
 | 
						|
    result = parse_terminal(parser)
 | 
						|
    while parser.match("*"):
 | 
						|
        op = parser.prev()
 | 
						|
        rhs = parse_terminal(parser)
 | 
						|
        result = ["*", result, rhs]
 | 
						|
    return result
 | 
						|
 | 
						|
def parse_terminal(parser):
 | 
						|
    # If we reach here, the current token *must* be a number.
 | 
						|
    return parser.consume()
 | 
						|
 | 
						|
def evaluate_ast(ast):
 | 
						|
    if type(ast) == int:
 | 
						|
        return ast
 | 
						|
    else:
 | 
						|
        op, lhs, rhs = ast[0], ast[1], ast[2]
 | 
						|
        return apply_operator(op, evaluate_ast(lhs), evaluate_ast(rhs))
 | 
						|
 | 
						|
def evaluate_recursive_descent(expr):
 | 
						|
    tokens = tokenize(expr)
 | 
						|
    print("Tokens:  {}".format(tokens))
 | 
						|
    ast = parse_expression(tokens)
 | 
						|
    print("AST:     {}".format(ast))
 | 
						|
    result = evaluate_ast(ast)
 | 
						|
    return result
 | 
						|
 | 
						|
methods = {
 | 
						|
    'Polish Notation': evaluate_with_polish_notation,
 | 
						|
    'Evaluate Tokens': evaluate_tokens,
 | 
						|
    'Ad Hoc Tree': evaluate_ad_hoc_tree,
 | 
						|
    'Recursive Descent': evaluate_recursive_descent,
 | 
						|
}
 | 
						|
 | 
						|
for name, fn in methods.items():
 | 
						|
    expr = "13 + 2 * 4 + 7 + 3 * 8"
 | 
						|
    print("Evaluating \"{}\" using the \"{}\" method...".format(expr, name))
 | 
						|
    assert fn(expr) == eval(expr)
 | 
						|
    print("Success!")
 |