def runall(cuda_device, model_desc, db_string):
labnotebook.initialize(db_string)
torch.cuda.set_device(cuda_device)
torch.manual_seed(model_desc['seed'])
model = MnistModel()
train_loader, test_loader = sgld.make_datasets()
model = model.cuda()
if model_desc['optimizer'][:4] == 'sgld':
optimizer = eval(model_desc['optimizer'])(
model.parameters(),
lr=model_desc['lr'],
addnoise=model_desc['addnoise'])
else:
optimizer = eval(model_desc['optimizer'])(model.parameters(),
lr=model_desc['lr'])
xp = train(model, train_loader, test_loader, optimizer, model_desc,
cuda_device)
return xp
def start_backend():
parser = argparse.ArgumentParser()
parser.add_argument("database_url")
parser.add_argument("--host", default="127.0.0.1")
args = parser.parse_args()
app = Flask(__name__)
# this is for the ability to run everything from different servers.
cors = CORS(app)
api = Api(app)
xp, ts, mp = labnotebook.initialize(args.database_url)
def flip_dict(dict_list):
"""
helper function to flip list of dicts to dict of lists
"""
result = defaultdict(list)
for dic in dict_list:
for k, v in dic.items():
result[k].append(v)
return result
class Steps(Resource):
def get(self, run_id):
query = labnotebook.session.query(
ts.step_id, ts.timestep, ts.run_id, ts.trainacc, ts.valacc,
ts.trainloss).filter(ts.run_id == run_id).all()
result = labnotebook.runs_schema.dump(query)[0]
return jsonify(flip_dict(result))
class Experiments(Resource):
def get(self):
query = labnotebook.session.query(
xp.run_id, xp.dt, xp.gpu, xp.completed, xp.final_trainacc,
xp.final_trainloss, xp.final_valacc,
xp.model_desc).order_by(xp.run_id.desc()).all()
result = labnotebook.xps_schema.dump(query)[0]
return jsonify(result)
class CustomFieldNames(Resource):
def get(self, run_id):
query = labnotebook.session.query(
func.jsonb_object_keys(ts.custom_fields)).filter(
ts.run_id == run_id).filter(ts.timestep == 1).all()
query = [q[0] for q in query]
return jsonify(query)
class CustomFields(Resource):
def get(self, run_id):
parser = reqparse.RequestParser()
parser.add_argument('fieldname', type=str)
fieldname = parser.parse_args()['fieldname']
query = labnotebook.session.query(
ts.timestep, ts.custom_fields[fieldname].label('cf')).filter(
ts.run_id == run_id).all()
result = labnotebook.cfs_schema.dump(query)[0]
return jsonify(flip_dict(result))
# our api:
api.add_resource(Steps, '/steps/<string:run_id>')
api.add_resource(CustomFields, '/customfields/<string:run_id>')
api.add_resource(CustomFieldNames, '/customfieldnames/<string:run_id>')
api.add_resource(Experiments, '/experiments')
app.run(debug=True, host=args.host, port=3000)
import labnotebook
from mas_tools.api import Binance
# logging
path = 'E:/Projects/market-analysis-system/mas_arbitrage/'
logging.basicConfig(level=logging.INFO,
handlers=[
logging.FileHandler("{p}/{fn}.log".format(
p=path, fn='bot_0.0')),
logging.StreamHandler()
])
log = logging.getLogger()
# labnotebook
db_url = 'postgres://*****:*****@localhost/postgres'
experiments, steps, model_params = labnotebook.initialize(db_url)
# cripto exchange api
MY_API_KEY = '---'
MY_API_SECRET = '---'
api = Binance(MY_API_KEY, MY_API_SECRET)
# parameters
symb1 = 'BTCUSDT'
symb2 = 'ETHUSDT'
period = '1m'
arbitrage_sum = True
coef = 15
h_level = 1345
l_level = 1180