def update_model(model):
fts, reward = get_cust_reward()
action = np.random.randint(3)
context = np.array(fts).reshape(-1)
reward = float(reward[action])
model.update(context, action, reward)
return model
def create_toy_contexual_dataset():
dataset = ContextualDataset(context_dim=2,
num_actions=3,
memory_size=-1,
intercept=False)
for i in range(2000):
fts, reward = get_cust_reward()
action = i % 3
r = reward[action]
dataset.add(fts, action, r)
return dataset
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')