File size: 9,186 Bytes
983012f c42a295 983012f c42a295 983012f c42a295 983012f c42a295 983012f c42a295 983012f 007ad5c 983012f 6f5abb0 983012f 1a23b39 983012f 1a23b39 983012f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 |
import gradio as gr
from gradio.components import Textbox, Slider, Plot
import numpy as np
import matplotlib.pyplot as plt
from tqdm.auto import tqdm
plt.style.use('seaborn-v0_8-whitegrid')
columns = ["difficulty", "stability", "retrievability", "delta_t",
"reps", "lapses", "last_date", "due", "ivl", "cost", "rand"]
col = {key: i for i, key in enumerate(columns)}
first_rating_prob = np.array([0.15, 0.2, 0.6, 0.05])
def simulate(w, request_retention=0.9, deck_size=10000, learn_span=100, max_cost_perday=200, max_ivl=36500, recall_cost=10, forget_cost=30, learn_cost=10):
card_table = np.zeros((len(columns), deck_size))
card_table[col["due"]] = learn_span
card_table[col["difficulty"]] = 1e-10
card_table[col["stability"]] = 1e-10
review_cnt_per_day = np.zeros(learn_span)
learn_cnt_per_day = np.zeros(learn_span)
memorized_cnt_per_day = np.zeros(learn_span)
def stability_after_success(s, r, d, response):
hard_penalty = np.where(response == 1, w[15], 1)
easy_bonus = np.where(response == 3, w[16], 1)
return s * (1 + np.exp(w[8]) * (11 - d) * np.power(s, -w[9]) * (np.exp((1 - r) * w[10]) - 1) * hard_penalty * easy_bonus)
def stability_after_failure(s, r, d):
return np.maximum(0.1, np.minimum(w[11] * np.power(d, -w[12]) * (np.power(s + 1, w[13]) - 1) * np.exp((1 - r) * w[14]), s))
for today in tqdm(range(learn_span)):
has_learned = card_table[col["stability"]] > 1e-10
card_table[col["delta_t"]][has_learned] = today - \
card_table[col["last_date"]][has_learned]
card_table[col["retrievability"]][has_learned] = np.power(
1 + card_table[col["delta_t"]][has_learned] / (9 * card_table[col["stability"]][has_learned]), -1)
card_table[col["cost"]] = 0
need_review = card_table[col["due"]] <= today
card_table[col["rand"]][need_review] = np.random.rand(
np.sum(need_review))
forget = card_table[col["rand"]] > card_table[col["retrievability"]]
card_table[col["cost"]][need_review & forget] = forget_cost
card_table[col["cost"]][need_review & ~forget] = recall_cost
true_review = need_review & (
np.cumsum(card_table[col["cost"]]) <= max_cost_perday)
card_table[col["last_date"]][true_review] = today
card_table[col["lapses"]][true_review & forget] += 1
card_table[col["reps"]][true_review & ~forget] += 1
card_table[col["stability"]][true_review & forget] = stability_after_failure(
card_table[col["stability"]][true_review & forget], card_table[col["retrievability"]][true_review & forget], card_table[col["difficulty"]][true_review & forget])
review_ratings = np.random.choice([1, 2, 3], np.sum(true_review & ~forget), p=[0.3, 0.6, 0.1])
card_table[col["stability"]][true_review & ~forget] = stability_after_success(
card_table[col["stability"]][true_review & ~forget], card_table[col["retrievability"]][true_review & ~forget], card_table[col["difficulty"]][true_review & ~forget], review_ratings)
card_table[col["difficulty"]][true_review & forget] = np.clip(
card_table[col["difficulty"]][true_review & forget] + 2 * w[6], 1, 10)
need_learn = card_table[col["due"]] == learn_span
card_table[col["cost"]][need_learn] = learn_cost
true_learn = need_learn & (
np.cumsum(card_table[col["cost"]]) <= max_cost_perday)
card_table[col["last_date"]][true_learn] = today
first_ratings = np.random.choice(4, np.sum(true_learn), p=first_rating_prob)
card_table[col["stability"]][true_learn] = np.choose(
first_ratings, w[:4])
card_table[col["difficulty"]][true_learn] = w[4] - \
w[5] * (first_ratings - 3)
card_table[col["ivl"]][true_review | true_learn] = np.clip(np.round(
9 * card_table[col["stability"]][true_review | true_learn] * (1 / request_retention - 1), 0), 1, max_ivl)
card_table[col["due"]][true_review | true_learn] = today + \
card_table[col["ivl"]][true_review | true_learn]
review_cnt_per_day[today] = np.sum(true_review)
learn_cnt_per_day[today] = np.sum(true_learn)
memorized_cnt_per_day[today] = card_table[col["retrievability"]].sum()
return card_table, review_cnt_per_day, learn_cnt_per_day, memorized_cnt_per_day
def interface_func(weights: str, learning_time: int, learn_span: int, deck_size: int, max_ivl: int, recall_cost: int, forget_cost: int, learn_cost: int,
progress=gr.Progress(track_tqdm=True)):
plt.close('all')
np.random.seed(42)
weights = weights.replace('[', '').replace(']', '')
w = list(map(lambda x: float(x.strip()), weights.split(',')))
max_cost_perday = learning_time * 60
def moving_average(data, window_size=learn_span//20):
weights = np.ones(window_size) / window_size
return np.convolve(data, weights, mode='valid')
for request_retention in [0.95, 0.9, 0.85, 0.8, 0.75]:
(_,
review_cnt_per_day,
learn_cnt_per_day,
memorized_cnt_per_day) = simulate(w,
request_retention=request_retention,
deck_size=deck_size,
learn_span=learn_span,
max_cost_perday=max_cost_perday,
max_ivl=max_ivl,
recall_cost=recall_cost,
forget_cost=forget_cost,
learn_cost=learn_cost)
plt.figure(1)
plt.plot(moving_average(review_cnt_per_day),
label=f"R={request_retention*100:.0f}%")
plt.title("Review Count per Day")
plt.legend()
plt.figure(2)
plt.plot(moving_average(learn_cnt_per_day),
label=f"R={request_retention*100:.0f}%")
plt.title("Learn Count per Day")
plt.legend()
plt.figure(3)
plt.plot(np.cumsum(learn_cnt_per_day),
label=f"R={request_retention*100:.0f}%")
plt.title("Cumulative Learn Count")
plt.legend()
plt.figure(4)
plt.plot(memorized_cnt_per_day,
label=f"R={request_retention*100:.0f}%")
plt.title("Memorized Count per Day")
plt.legend()
return plt.figure(1), plt.figure(2), plt.figure(3), plt.figure(4)
description = f"""
# FSRS4Anki Simulator
Here is a simulator for FSRS4Anki. It can simulate the learning process of a deck with given weights and parameters.
It will help you to find the expected requestRetention for FSRS4Anki.
The simulator assumes that you spend the same amount of time on Anki every day.
"""
with gr.Blocks() as demo:
with gr.Group():
gr.Markdown(description)
with gr.Group():
with gr.Row():
with gr.Column():
weights = Textbox(label="Weights", lines=1,
value="0.4, 0.6, 2.4, 5.8, 4.93, 0.94, 0.86, 0.01, 1.49, 0.14, 0.94, 2.18, 0.05, 0.34, 1.26, 0.29, 2.61")
learning_time = Slider(label="Learning Time perday (minutes)",
minimum=5, maximum=1440, step=5, value=30)
learn_span = Slider(label="Learning Period (days)", minimum=30,
maximum=3650, step=10, value=365)
deck_size = Slider(label="Deck Size (cards)", minimum=100,
maximum=100000, step=100, value=10000)
with gr.Column():
max_ivl = Slider(label="Maximum Interval (days)", minimum=30,
maximum=36500, step=10, value=36500)
recall_cost = Slider(label="Review Cost (seconds)", minimum=1,
maximum=600, step=1, value=10)
forget_cost = Slider(label="Relearn Cost (seconds)",
minimum=1, maximum=600, step=1, value=30)
learn_cost = Slider(label="Learn Cost (seconds)", minimum=1,
maximum=600, step=1, value=10)
with gr.Row():
btn_plot = gr.Button("Simulate")
with gr.Row():
with gr.Column():
review_count = Plot(label="Review Count per Day")
learn_count = Plot(label="Learn Count per Day")
with gr.Column():
cumulative_learn_count = Plot(label="Cumulative Learn Count")
memorized_count = Plot(label="Memorized Count per Day")
btn_plot.click(
fn=interface_func,
inputs=[weights,
learning_time,
learn_span,
deck_size,
max_ivl,
recall_cost,
forget_cost,
learn_cost,
],
outputs=[review_count, learn_count,
cumulative_learn_count, memorized_count],
)
demo.queue().launch(show_error=True)
|