def fit_model(model):
data = generate_dataframe(220)
ctx = data[['age', 'ARPU']]
actions = data['action']
rewards = data['reward']
model.fit(ctx, actions, rewards)
return model
def test_update(self):
model = LinearBandits(
3,
4
)
context = np.array([76, 120, 654326, 2])
action = 1
reward = 14
model.update(context, action, reward)
df = generate_dataframe(500)
contexts = df[['age', 'ARPU']]
actions = df['action']
rewards = df['reward']
new_model = NeuralBandits(3, 2, layer_sizes=[50, 12], verbose=False)
#call .fit method; num_updates will repeat training n times
new_model.fit(contexts, actions, rewards)
new_context = np.array([26.0, 98.456463])
action = np.random.randint(0, 2)
reward = np.random.random() * 10
print(action)
new_model.update(new_context, action, reward)
def check_toy_problem():
customer = get_customer()
ctype, (age, ft) = customer
assert isinstance(ctype, int)
assert isinstance(age, int)
assert isinstance(ft, float)
reward = get_rewards(customer)
assert reward.shape == (3, )
fts, reward = get_cust_reward()
df = generate_dataframe(10)
assert isinstance(df, pd.DataFrame)
return fts, reward
def test_neural_linear_model(self):
model = NeuralBandits(num_actions=3,
num_features=2,
training_freq_network=200,
layer_sizes=[50])
fts, reward = get_cust_reward()
for i in range(300):
action = model.action(fts)
r = reward[action]
model.update(fts, action, r)
df = generate_dataframe(500)
X = df[['age', 'ARPU']].values
A = df['action'].values
R = df['reward'].values
model.fit(X, A, R)
model.save('test_file')
model = load_model('test_file')
X = df[['age', 'ARPU']].sample(2).values
model.predict(X, parallelize=False)
os.remove('test_file')
def predict_model(model):
data = generate_dataframe(220)
ctx = data[['age', 'ARPU']]
predictions = model.predict(ctx)
return predictions