def fit():
# X_file = request.files['X']
# y_file = request.files['y']
if not request.is_json:
return "NO JSON!"
data = request.json
X = np.array(data['X'])
y = np.array(data['y'])
# X = np.array(json.loads(X_file.read().decode('utf-8')))
# y = np.array(json.loads(y_file.read().decode('utf-8')))
model = RegressionModel()
model.fit(X, y)
model_bytes = pickle.dumps(model)
model_ddb.add_model_to_db(dynamodb,
models_table_name,
model_bytes)
data_ddb.add_fit_data_to_db(dynamodb,
data_table_name,
pickle.dumps(X),
pickle.dumps(y))
request_url = request.url
request_ddb.add_request_to_db(dynamodb,
requests_table_name,
request_url)
return 'ok'
def main():
num_inducing = args.num_inducing
data_name = args.fname
prop = args.prop
n_layers = args.layers
with open("settings-uci.pkl", "rb") as file:
settings = pickle.load(file)
data = pd.read_csv('./datasets/{}.csv'.format(data_name),
header=None).values
key = data_name + "-{}".format(n_layers)
try:
adam_lr = [settings[key][0], settings[key][1], settings[key][2]]
max_iter = settings[key][3]
except:
adam_lr = [0.005, 0.0001, 0.0025]
max_iter = 20000
if data_name == "energy":
X_full = data[:, :-2]
Y_full = data[:, -2:-1]
else:
X_full = data[:, :-1]
Y_full = data[:, -1:]
N = X_full.shape[0]
n = int(N * prop)
np.random.seed(0)
ind = np.arange(N)
np.random.shuffle(ind)
train_ind = ind[:n]
test_ind = ind[n:]
X = X_full[train_ind]
Xs = X_full[test_ind]
Y = Y_full[train_ind]
Ys = Y_full[test_ind]
X_mean = np.mean(X, 0)
X_std = np.std(X, 0)
Y_std = np.std(Y, 0)
X = (X - X_mean) / X_std
Xs = (Xs - X_mean) / X_std
Y_mean = np.mean(Y, 0)
Y = (Y - Y_mean) / Y_std
Ys = (Ys - Y_mean) / Y_std
model = RegressionModel(adam_lr, max_iter, n_layers, num_inducing)
model.fit(X, Y, Xs, Ys, Y_std)
def main(X_filename='data/X_train.json',
y_filename='data/y_train.json',
model_filename='regression_model.pkl'):
"""
Init RegressionModel object and call `fit` method with corresponding arguments and save it as `pickle`.
**Parameters**:
- `X_filename`: filename where object-feature matrix is saved
- `y_filename`: filename of `y` values form `f(X) = y`
- `model_filename`: name of file to save model
"""
X = np.array(json.load(codecs.open(X_filename, 'r', encoding='utf-8')))
y = np.array(json.load(codecs.open(y_filename, 'r', encoding='utf-8')))
reg_model = RegressionModel()
reg_model.fit(X, y)
with open(model_filename, 'wb') as f:
pickle.dump(reg_model, f)