update the various comet scripts

DCNN-Pytorch/get_comet_experiment_ame2e.py

@@ -17,8 +17,9 @@
 parser.add_argument("--epoch_number", type=int, required=False, default=100, help="Experiment key to grab from comet.")
 parser.add_argument("--output_directory", type=str, required=False, default=".", help="Where to put the config and data files.")
 
+projectdict = {"cnnlstm" : "deepracingcnnlstm", "bezier" : "deepracingbezierpredictor", "pilot" : "deepracingpilotnet"}
 #XMUs9uI19KQNdYrQhuXPnAfpB
-args = parser.parse_args() 
+args = parser.parse_args()
 experiment_key = args.experiment_key
 epoch_number = args.epoch_number
 restkey = args.restkey
@@ -29,60 +30,50 @@
     api = API()
 else:
     api = API(rest_api_key=restkey)
-
-experiment : APIExperiment = api.get_experiment("electric-turtle", "deepracingadmiralnet-e2e", experiment_key)
+experiment : APIExperiment = api.get_experiment("electric-turtle", "deepracingposepredictor", experiment_key)
 assetlist = experiment.get_asset_list()
 assetdict = {d['fileName']: d['assetId'] for d in assetlist}
 
-
-print("Getting hyperparameters from comet")
-parameters_summary = experiment.get_parameters_summary()
-configin = {d["name"] : d["valueCurrent"] for d in parameters_summary}
-config = {}
-config["image_size"] = np.fromstring( configin["image_size"].replace(" ","").replace("[","").replace("]",""), sep=',', dtype=np.int32 ).tolist()
-# try:
-#     config["image_size"] = np.fromstring( configin["image_size"].replace(" ","").replace("[","").replace("]",""), sep=',', dtype=np.int32 ).tolist()
-# except:
-#     config["image_size"] = [66,200]
-config["input_channels"] = int( configin["input_channels"] )
-config["hidden_dimension"] = int( configin["hidden_dimension"] )
-config["sequence_length"] = int( configin["sequence_length"] )
-config["context_length"] = int( configin["context_length"] )
-config["output_dimension"] = int( configin["output_dimension"] )
-
-print(config)
-outputconfigfile = os.path.join(output_directory,"config.yaml")
-with open(outputconfigfile, 'w') as f:
-    yaml.dump(config,stream=f,Dumper=yaml.SafeDumper)
-
-
 #get network weight file
-weightfilename = "admiralnet_epoch_%d_params.pt" %(epoch_number,)
-optimizerfilename = "admiralnet_epoch_%d_optimizer.pt" %(epoch_number,)
+weightfilename = "epoch_%d_params.pt" %(epoch_number,)
+optimizerfilename = "epoch_%d_optimizer.pt" %(epoch_number,)
 
 print("Getting network weights from comet")
 params_binary = experiment.get_asset(assetdict[weightfilename])
 
-outputweightfile = os.path.join(output_directory,weightfilename)
-with open(outputweightfile, 'wb') as f:
-    f.write(params_binary)
-
 #get optimizer weight file
 print("Getting optimizer weights from comet")
 optimizer_binary = experiment.get_asset(assetdict[optimizerfilename])
 
+outputweightfile = os.path.join(output_directory,weightfilename)
+with open(outputweightfile, 'wb') as f:
+    f.write(params_binary)
 
 outputoptimizerfile = os.path.join(output_directory,optimizerfilename)
 with open(outputoptimizerfile, 'wb') as f:
     f.write(optimizer_binary)
 
+#get parameters
+parameters_summary = experiment.get_parameters_summary()
+configin = {d["name"] : d["valueCurrent"] for d in parameters_summary}
+config = {}
+config["image_size"] = np.fromstring(configin["image_size"].replace(" ","")[1:-1],sep=',').astype(np.int32).tolist()
+config["input_channels"] = int( configin["input_channels"] )
+config["hidden_dimension"] = int( configin["hidden_dimension"] )
+config["sequence_length"] = int( configin["sequence_length"] )
+config["context_length"] = int( configin["context_length"] )
+print(config)
+
+
+outputconfigfile = os.path.join(output_directory,"config.yaml")
+with open(outputconfigfile, 'w') as f:
+    yaml.dump(config,stream=f,Dumper=yaml.SafeDumper)
 
 context_length = config["context_length"]
-sequence_length = config["sequence_length"]
 input_channels = config["input_channels"]
 hidden_dimension = config["hidden_dimension"]
-output_dimension = config["output_dimension"]
-net = deepracing_models.nn_models.Models.AdmiralNetKinematicPredictor( context_length = context_length , sequence_length = sequence_length, input_channels=input_channels, hidden_dim = hidden_dimension, output_dimension=output_dimension) 
+sequence_length = config["sequence_length"]
+net = deepracing_models.nn_models.Models.AdmiralNetKinematicPredictor( sequence_length = sequence_length, context_length = context_length , input_channels=input_channels, hidden_dim = hidden_dimension ) 
 net = net.double()
 with open(outputweightfile, 'rb') as f:
     net.load_state_dict(torch.load(f, map_location=torch.device("cpu")))

DCNN-Pytorch/get_comet_experiment_cnnlstm.py

@@ -7,9 +7,7 @@
 import torch.nn as NN
 import numpy as np
 import torch.utils.data as data_utils
-import deepracing_models.data_loading.proto_datasets as PD
 from tqdm import tqdm as tqdm
-import deepracing_models.nn_models.LossFunctions as loss_functions
 import deepracing_models.nn_models.Models
 parser = argparse.ArgumentParser(description="Download experiment from Comet.ml")
 parser.add_argument("experiment_key", type=str, help="Experiment key to grab from comet.")
@@ -22,7 +20,7 @@
 experiment_key = args.experiment_key
 epoch_number = args.epoch_number
 restkey = args.restkey
-output_directory = os.path.join(args.output_directory, experiment_key + "_from_comet")
+output_directory = os.path.join(args.output_directory, experiment_key)
 if not os.path.isdir(output_directory):
     os.makedirs(output_directory)
 if restkey is None:
@@ -34,16 +32,19 @@
 assetdict = {d['fileName']: d['assetId'] for d in assetlist}
 #print(assetdict)
 #get network weight file
-weightfilename = "pilotnet_epoch_%d_params.pt" %(epoch_number,)
-weightfilenamealt = "CNNLSTM_epoch_%d_params.pt" %(epoch_number,)
+weightfilename = "CNNLSTM_epoch_%d_params.pt" %(epoch_number,)
+weightfilenamealt = "epoch_%d_params.pt" %(epoch_number,)
+weightfilenamealtalt = "pilotnet_epoch_%d_params.pt" %(epoch_number,)
 
 print("Getting network weights from comet")
 try:
     params_binary = experiment.get_asset(assetdict[weightfilename])
-    outputweightfile = os.path.join(output_directory,weightfilename)
 except KeyError as e:
-    params_binary = experiment.get_asset(assetdict[weightfilenamealt])
-    outputweightfile = os.path.join(output_directory,weightfilenamealt)
+    try:
+        params_binary = experiment.get_asset(assetdict[weightfilename])
+    except KeyError as e:
+        params_binary = experiment.get_asset(assetdict[weightfilenamealtalt])
+outputweightfile = os.path.join(output_directory,"epoch_%d_params.pt" %(epoch_number,))
 
 
 
@@ -56,7 +57,7 @@
 config = {}
 config["image_size"] = np.fromstring( configin["image_size"].replace(" ","")[1:-1], sep=',', dtype=np.int32).tolist()
 config["input_channels"] = int( configin["input_channels"] )
-config["output_dimension"] = int( configin["output_dimension"] )
+config["output_dimension"] = int( configin.get("output_dimension","2") )
 config["hidden_dimension"] = int( configin.get("hidden_dimension","100") )
 config["sequence_length"] = int( configin["sequence_length"] )
 config["context_length"] = int( configin.get("context_length","5") )

DCNN-Pytorch/get_comet_experiment_waypoint.py

@@ -7,23 +7,23 @@
 import torch.nn as NN
 import numpy as np
 import torch.utils.data as data_utils
-import deepracing_models.data_loading.proto_datasets as PD
 from tqdm import tqdm as tqdm
-import deepracing_models.nn_models.LossFunctions as loss_functions
 import deepracing_models.nn_models.Models
 parser = argparse.ArgumentParser(description="Download experiment from Comet.ml")
 parser.add_argument("experiment_key", type=str, help="Experiment key to grab from comet.")
 parser.add_argument("--restkey", type=str, required=False, default=None, help="Experiment key to grab from comet.")
 parser.add_argument("--epoch_number", type=int, required=False, default=100, help="Experiment key to grab from comet.")
 parser.add_argument("--output_directory", type=str, required=False, default=".", help="Where to put the config and data files.")
+parser.add_argument("--get_optimizer_weights", action="store_true", help="Also grab the state dictionary of the optimizer at the end of the specified epoch")
 
 projectdict = {"cnnlstm" : "deepracingcnnlstm", "bezier" : "deepracingbezierpredictor", "pilot" : "deepracingpilotnet"}
 #XMUs9uI19KQNdYrQhuXPnAfpB
 args = parser.parse_args()
 experiment_key = args.experiment_key
 epoch_number = args.epoch_number
 restkey = args.restkey
-output_directory = os.path.join(args.output_directory, experiment_key + "_from_comet")
+get_optimizer_weights = args.get_optimizer_weights
+output_directory = os.path.join(args.output_directory, experiment_key)
 if not os.path.isdir(output_directory):
     os.makedirs(output_directory)
 if restkey is None:
@@ -36,22 +36,20 @@
 
 #get network weight file
 weightfilename = "epoch_%d_params.pt" %(epoch_number,)
-optimizerfilename = "epoch_%d_optimizer.pt" %(epoch_number,)
-
 print("Getting network weights from comet")
 params_binary = experiment.get_asset(assetdict[weightfilename])
-
-#get optimizer weight file
-print("Getting optimizer weights from comet")
-optimizer_binary = experiment.get_asset(assetdict[optimizerfilename])
-
 outputweightfile = os.path.join(output_directory,weightfilename)
 with open(outputweightfile, 'wb') as f:
     f.write(params_binary)
 
-outputoptimizerfile = os.path.join(output_directory,optimizerfilename)
-with open(outputoptimizerfile, 'wb') as f:
-    f.write(optimizer_binary)
+if get_optimizer_weights:
+    #get optimizer weight file
+    print("Getting optimizer weights from comet")
+    optimizerfilename = "epoch_%d_optimizer.pt" %(epoch_number,)
+    optimizer_binary = experiment.get_asset(assetdict[optimizerfilename])
+    outputoptimizerfile = os.path.join(output_directory,optimizerfilename)
+    with open(outputoptimizerfile, 'wb') as f:
+        f.write(optimizer_binary)
 
 #get parameters
 parameters_summary = experiment.get_parameters_summary()