def downloadArtifact(proj, exp_name, exp_run, serialization): client = Client("http://localhost:3000") proj = client.set_project(proj) expt = client.set_experiment(exp_name) run = client.set_experiment_run(exp_run) if serialization.lower() == 'pickle': run.download_model('model.pkl')
from verta import Client import cloudpickle client = Client('https://dev.verta.ai') proj = client.set_project('Demo - Jenkins+Prometheus') run = proj.expt_runs[0] model = run.get_model() with open('model.pkl', 'wb') as f: cloudpickle.dump(model, f)
import os import random import multiprocessing from verta import Client from verta.utils import ModelAPI os.environ['VERTA_EMAIL'] = '*****@*****.**' os.environ['VERTA_DEV_KEY'] = '3e078522-e479-4cd2-b78c-04ffcacae3f4' HOST = "dev.verta.ai" EXPERIMENT_NAME = "Scaling" client = Client(HOST) proj = client.set_project('Scaling Test 100 jobs of 500k models') expt = client.set_experiment(EXPERIMENT_NAME) # Hyperparam random choice of values c_list = [0.0001, 0.0002, 0.0004] solver_list = ['lgfgs', 'grad'] max_iter_list = [7, 15, 28] # results into 30 metric or hyp keys paramKeyLimit = 10 def getMetrics(key_limit): metric_obj = {} for i in range(key_limit): metric_obj['val_acc' + str(i)] = random.uniform(0.5, 0.9) metric_obj['loss' + str(i)] = random.uniform(0.6, 0.8)
from verta import Client from verta.utils import ModelAPI HOST = "http://localhost:3009" PROJECT_NAME = "readmission_shared_data_preprocess_v0" EXPERIMENT_NAME = "readmission_shared_data_preprocess_v0_first_run" if __name__ == '__main__': parser = argparse.ArgumentParser(description='Template Main Function') parser.add_argument('--input', type=str, help='input folder') parser.add_argument('--output', type=str, help='output folder') parser.add_argument('--vis', type=str, help='vis folder') args = parser.parse_args() client = Client(HOST) proj = client.set_project(PROJECT_NAME) expt = client.set_experiment(EXPERIMENT_NAME) run = client.set_experiment_run() cmd = "readmission_shared_data_preprocess_trainvalidationtest.sh" # modify lib_param = {} lib_param["--input"] = args.input lib_param["--output"] = args.output lib_param["--vis"] = args.vis for k, v in lib_param.items(): cmd = cmd + " " + str(k) + " " + str(v) cmd = "bash " + cmd print("executing cmd: \n", cmd) os.system(cmd) # modify
from nntoolbox.metrics import Accuracy, Loss from nntoolbox.losses import SmoothedCrossEntropy from verta import Client from verta.client import ExperimentRun from experii.verta import ModelDBCB from experii.ax import AxTuner torch.backends.cudnn.benchmark = True EXPERIMENT_NAME = "Hyperparameter Tuning" # Set up ModelDB: client = Client( CLIENT_PARA ) # supply your own ModelDB'S client parameters here (see VertaAI's notebooks) proj = client.set_project("My second ModelDB project") exp = client.set_experiment(EXPERIMENT_NAME) # Define model generating function: def model_fn(parameterization: Dict[str, Any]) -> nn.Module: model = Sequential( ConvolutionalLayer(in_channels=3, out_channels=16, kernel_size=3, activation=nn.ReLU), ResidualBlockPreActivation(in_channels=16, activation=nn.ReLU), ConvolutionalLayer(in_channels=16, out_channels=32, kernel_size=3, activation=nn.ReLU),