def model_learning(prediction_model, time_model, data, length=200):
enums = defaultdict(lambda: 0)
model = TimeCombiner(prediction_model, time_model)
model.pre_process_data(data)
df = []
for answer in data.iter():
item = answer["item"]
student = answer["student"]
prediction, time_prediction = model.predict(student, item, answer)
model.update(student, item, prediction, time_prediction, answer["correct"], answer["response_time"], answer)
skill = model._prediction_model.get_skill(student)
speed = model._time_model.get_skill(student)
enums[student] += 1
df.append([item, student, skill, speed, enums[student],
answer["correct"], np.log(answer["response_time"]),
prediction, np.log(time_prediction),
model._prediction_model.get_difficulty(item),
model._time_model.get_difficulty(item),
])
df = pd.DataFrame(df, columns=('item', 'student', 'skill', 'speed', 'enum', 'correct', 'response_time_log', 'prediction', 'time_prediction_log', 'difficulty', 'intensity'))
points = range(1, length + 1)
plt.subplot(321)
plt.title(str(model))
plt.plot(points, [df.loc[df['enum'] == p, 'skill'].mean() for p in points])
plt.xlabel('# answer')
plt.ylabel('skill')
plt.subplot(323)
plt.plot(points, [df.loc[df['enum'] == p, 'speed'].mean() for p in points])
plt.xlabel('# answer')
plt.ylabel('speed')
plt.subplot(325)
plt.bar(points[:-1], [(df['enum'] == p).sum() for p in points[:-1]])
plt.ylabel('User count')
plt.xlabel('# answer')
plt.subplot(322)
plt.plot(points, [df.loc[df['enum'] == p, 'correct'].mean() for p in points], label='observation')
plt.plot(points, [df.loc[df['enum'] == p, 'prediction'].mean() for p in points], label='prediction')
plt.legend(loc=3)
plt.xlabel('# answer')
plt.ylabel('success')
plt.subplot(324)
plt.plot(points, [np.exp(df.loc[df['enum'] == p, 'response_time_log'].mean()) for p in points], label='observation')
plt.plot(points, [np.exp(df.loc[df['enum'] == p, 'time_prediction_log'].mean()) for p in points], label='prediction')
plt.legend(loc=3)
plt.xlabel('# answer')
plt.ylabel('time')
plt.subplot(326)
plt.plot(points, [df.loc[df['enum'] == p, 'intensity'].mean() for p in points])
plt.ylabel('Intensity')
# plt.plot(points, [df.loc[df['enum'] == p, 'difficulty'].mean() for p in points])
# plt.ylabel('Difficulty')
plt.xlabel('# answer')
df.to_pickle('model_learning.pd')
return df
from utils.runner import Runner
from utils.utils import grid_search, enumerate_df
def moving_average(a, n=3) :
ret = np.cumsum(a, dtype=float)
ret[n:] = ret[n:] - ret[:-n]
return ret[n - 1:] / n
data_all = d.Data("../data/matmat/2016-06-27/answers.pd")
data_all.trim_times()
data_long = d.Data("../data/matmat/2016-06-27/answers.pd", filter=(100, 100))
data_long.trim_times()
data = data_all
time_model = BasicTimeModel(alpha=0.6, beta=0.1, K=0.25)
model = TimeCombiner(AvgModel(), time_model)
skills = defaultdict(lambda: [])
def update(student, item):
skills[student].append(time_model.get_skill(student))
model.after_update_callback = update
Runner(data, model).run(force=True)
# skills = time_model.get_skills(data.get_students())
# sns.distplot(skills)
length = 200
