Scott Aaronson Oracle

spike-aaronson-oracle.py

# %%
import collections
import random


def predict_next_letter(model, five_gram):
    if len(five_gram) != 5:
        raise ValueError("five_gram must be of length 5")
    m = model[tuple(five_gram)]
    return m[True] > m[False]


def train_fivegrams(input_s):
    model = collections.defaultdict(lambda: [0, 0])
    # Generate all six-grams from the input sequence
    six_grams = [input_s[i : i + 6] for i in range(len(input_s) - 5)]
    for s in six_grams:
        five_gram = s[:5]
        last = s[-1]
        model[tuple(five_gram)][last] += 1
    return model


# %%
def round(num):
    return int(num * 100)


def predict(model, press_q):
    if len(press_q) < 5:
        raise ValueError("press_q must be of length 5 or greater")
    else:
        five_gram = press_q[-5:]
        return predict_next_letter(model, five_gram)


press_q = []
correct_predictions = 0

print("Press 'f' or 'd' (or 'q' for quit and 'm' for model): ")

while True:
    model = train_fivegrams(press_q)  # Retrain the model with the history

    # Also draw a random number to simulate the user's behavior (50:50)
    # press = "f" if random.random() > 0.5 else "d"
    press = input()
    if press in ["f", "d"]:
        yes_no = True if press == "f" else False

        if len(press_q) > 20:
            predicted_yes_no = predict(model, press_q)
            correct_predictions += predicted_yes_no == yes_no
            accuracy = correct_predictions / (len(press_q) - 20)
            prediction = "f" if predicted_yes_no else "d"
            print(f"Predicted: {prediction}, Accuracy: {round(accuracy)}%")
        else:
            print("Learning...")

        press_q.append(yes_no)
    if press == "q":
        break
    if press == "m":
        print(model)
# %%