Changes to run SJ jet training

CPLUOS · Dec 11, 2017 · 7ad07ce · 7ad07ce
1 parent 49a03ac
commit 7ad07ce
Show file tree

Hide file tree

Showing 18 changed files with 122 additions and 83 deletions.
diff --git a/SJ-Keras/Proj-ChannelImportance/training.py b/SJ-Keras/Proj-ChannelImportance/training.py
@@ -2,7 +2,7 @@
 from __future__ import division
 from __future__ import print_function
 
-import sys.path
+import sys
 import os.path
 
 import matplotlib
@@ -21,7 +21,7 @@
 
 import keras.backend as K
 
-from pipeline import DataLodaer
+from pipeline import DataLoader
 
 sys.path.append("..")
 from keras4jet.models.image.resnet import build_a_model
@@ -40,7 +40,7 @@ def main():
     parser.add_argument(
         '--log_dir', type=str,
 	    default='./logs/{name}-{date}'.format(
-        name="{name}", date=datetime.today().strftime("%Y-%m-%d_%H-%M-%S")),
+        name="name", date=datetime.today().strftime("%Y-%m-%d_%H-%M-%S")))
 
     parser.add_argument("--num_epochs", type=int, default=10)
     parser.add_argument("--num_gpus", type=int, default=len(get_available_gpus()))
@@ -59,18 +59,19 @@ def main():
 
     args = parser.parse_args()
 
-    data_dir = "../Data/FastSim_pt_100_500"
+    data_dir = "../../SJ-JetImage/images33x33/"
     if args.train_data == "dijet":
         train_data = os.path.join(data_dir, "")
     elif args.train_data == "zjet":
+        train_data = os.path.join(data_dir, "")
     else:
         raise ValueError("")
 
     log_dir = get_log_dir(
         path=args.log_dir.format(name=args.channel),
         creation=True)
 
-    logger = Logger(dpath=log_dir.path, "WRITE")
+    logger = Logger(dpath=log_dir.path, mode="WRITE")
     logger.get_args(args)
     logger["train_data"] = train_data
     logger["val_zjet_data"] = val_dijet_data
@@ -99,7 +100,7 @@ def main():
         cyclic=True)
 
     # build a model
-    _model = build_a_model(input_shape
+    _model = build_a_model(input_shape)
     model = multi_gpu_model(_model, gpus=args.num_gpus) 
 
 
@@ -187,14 +188,14 @@ def main():
         x="step",
         ys=[(tr_acc_, "Training Acc on {}"),
             ("val_acc_dijet", "Validation on Dijet"),
-            ("val_acc_zjet", "Validation on Z+jet")]
+            ("val_acc_zjet", "Validation on Z+jet")],
         title="Accuracy", xaxis="Step", yaxis="Accuracy")
 
     meter.prepare(
         x="step",
         ys=[(tr_loss_name, "Training Loss os {}"),
             ("val_acc_dijet", "Validation Loss on Dijet"),
-            ("val_acc_zjet", "Validation Loss on Z+jet")]
+            ("val_acc_zjet", "Validation Loss on Z+jet")],
         title="Loss", xaxis="Step", yaxis="Accuracy")
 
     meter.finish()

diff --git a/SJ-Keras/Proj-JetImage/models/.ipynb_checkpoints/nb_dev_capsnet-checkpoint.ipynb b/SJ-Keras/Proj-JetImage/models/.ipynb_checkpoints/nb_dev_capsnet-checkpoint.ipynb
diff --git a/SJ-Keras/Proj-JetImage/models/__init__.py b/SJ-Keras/Proj-JetImage/models/__init__.py
@@ -2,13 +2,13 @@
 from __future__ import division
 from __future__ import print_function
 
-import models.vggnet
-import models.resnet
-import models.densenet
-import models.seresnet
-import models.sedensnet
+# import models.image.vggnet
+# import models.resnet
+# import models.densenet
+# import models.seresnet
+# import models.sedensnet
 
-import models.capsnet
+# import models.capsnet
 
 __all__ = []
 

diff --git a/SJ-Keras/Proj-JetImage/pipeline.py b/SJ-Keras/Proj-JetImage/pipeline.py
@@ -15,6 +15,9 @@ def __init__(self, path, batch_size, cyclic, image_shape=(3, 33, 33)):
 
         self.root_file = ROOT.TFile(path, "READ")
         self.tree = self.root_file.Get("jet")
+        if self.tree == ROOT.MakeNullPointer():
+            print("ERROR:", self.root_file, "contains no 'jet' tree")
+            exit(1)
 
         self._start = 0
 

diff --git a/SJ-Keras/Proj-JetImage/training.py b/SJ-Keras/Proj-JetImage/training.py
@@ -21,11 +21,10 @@
 
 import keras.backend as K
 
-from pipeline import DataLodaer
-
-from models import build_a_model
+from pipeline import DataLoader
 
 sys.path.append("..")
+from keras4jet.models import build_a_model
 from keras4jet.meters import Meter
 from keras4jet.utils import (
     get_log_dir,
@@ -37,14 +36,18 @@ def main():
     parser = argparse.ArgumentParser()
 
     parser.add_argument("--train_data", type=str, default="dijet")
-	'--log_dir', type=str,
-	default='./logs/{name}-{date}'.format(
-        name="{name}", date=datetime.today().strftime("%Y-%m-%d_%H-%M-%S")),
+    parser.add_argument("--directory", type=str, default="../../SJ-JetImage/image33x33/")
+    parser.add_argument("--model", type=str, default="resnet")
+    parser.add_argument('--log_dir', type=str,
+	                default='./logs/{name}-{date}'.format(
+                            name="{name}",
+                            date=datetime.today().strftime("%Y-%m-%d_%H-%M-%S")))
 
     parser.add_argument("--num_epochs", type=int, default=10)
     parser.add_argument("--num_gpus", type=int, default=len(get_available_gpus()))
     parser.add_argument("--train_batch_size", type=int, default=500)
     parser.add_argument("--val_batch_size", type=int, default=500)
+    parser.add_argument("--multi-gpu", default=False, action='store_true', dest='multi_gpu')
 
     # Hyperparameter
     parser.add_argument("--lr", type=float, default=0.001)
@@ -56,39 +59,40 @@ def main():
     args = parser.parse_args()
 
     if args.train_data == "dijet":
-        train_data = "../data/FastSim/dijet/training_dijet_466554_prep.root"
-        val_dijet_data = "../data/FastSim/dijet/test_dijet_after_dijet_186880_prep.root"
-        val_zjet_data = "../data/FastSim/dijet/test_zjet_after_dijet_176773_prep.root"
+        train_data = args.directory+"/dijet_train.root"
+        val_dijet_data = args.directory+"/dijet_test_after_dijet.root"
+        val_zjet_data = args.directory+"/z_jet_test_after_dijet.root"
     elif args.train_data == "zjet":
-        train_data = "../data/FastSim/zjet/training_zjet_440168_prep.root"
-        val_dijet_data = "../data/FastSim/zjet/test_dijet_after_zjet_186880_prep.root"
-        val_zjet_data = "../data/FastSim/zjet/test_zjet_after_zjet_176773_prep.root"
+        train_data = args.directory+"/z_jet_train.root"
+        val_dijet_data = args.directory+"/dijet_test_after_zjet.root"
+        val_zjet_data = args.directory+"/z_jet_test_after_zjet.root"
     else:
         raise ValueError("")
 
+    if '{name}' in args.log_dir: args.log_dir = args.log_dir.format(name=args.model)
     log_dir = get_log_dir(path=args.log_dir, creation=True)
 
-    logger = Logger(dpath=log_dir.path, "WRITE")
+    logger = Logger(dpath=log_dir.path, mode="WRITE")
     logger.get_args(args)
     logger["train_data"] = train_data
-    logger["val_zjet_data"] = val_dijet_data
+    logger["val_zjet_data"] = val_zjet_data
     logger["val_dijet_data"] = val_dijet_data
 
     # data loader
-    train_loader = DataLodaer(
+    train_loader = DataLoader(
         path=train_data,
         batch_size=args.train_batch_size,
         cyclic=False)
 
     steps_per_epoch = np.ceil( len(train_loader) / train_loader.batch_size ).astype(int)
     total_step = args.num_epochs * steps_per_epoch
 
-    val_dijet_loader = DataLodaer(
+    val_dijet_loader = DataLoader(
         path=val_dijet_data,
         batch_size=args.val_batch_size,
         cyclic=True)
 
-    val_zjet_loader = DataLodaer(
+    val_zjet_loader = DataLoader(
         path=val_zjet_data,
         batch_size=args.val_batch_size,
         cyclic=True)
@@ -99,11 +103,12 @@ def main():
     metric_list = ['accuracy']
 
     # build a model and compile it
-    _model = build_a_model(
-        model_name=args.model,
-        **train_loader.get_shapes())
+    _model = build_a_model(model_name=args.model, input_shape=train_loader._image_shape)
 
-    model = multi_gpu_model(_model, gpus=args.num_gpus) 
+    if args.multi_gpu:
+        model = multi_gpu_model(_model, gpus=args.num_gpus)
+    else:
+        model = _model
 
 
     model.compile(
@@ -191,6 +196,9 @@ def main():
             model.train_on_batch(x=x_train, y=y_train)
             step += 1
 
+    filepath = os.path.join(log_dir.saved_models.path,
+                            "model_final.h5")
+    _model.save(filepath)
     print("Training is over! :D")
     meter.finish()
     logger.finish()

diff --git a/SJ-Keras/Proj-RNN/pipeline.py b/SJ-Keras/Proj-RNN/pipeline.py
@@ -25,6 +25,9 @@ def __init__(self, path, batch_size, cyclic, maxlen=54, tree_name="jetAnalyser")
 
         self.root_file = ROOT.TFile(path, "READ")
         self.tree = self.root_file.Get(tree_name)
+        if self.tree == ROOT.MakeNullPointer():
+            print("ERROR:", self.root_file, "contains no 'jet' tree")
+            exit(1)
 
         if maxlen is None:
             self.maxlen = int(self.tree.GetMaximum("n_dau"))
@@ -180,6 +183,9 @@ def __init__(self, path, batch_size, cyclic, maxlen=54, tree_name="jetAnalyser")
 
         self.root_file = ROOT.TFile(path, "READ")
         self.tree = self.root_file.Get(tree_name)
+        if self.tree == ROOT.MakeNullPointer():
+            print("ERROR:", self.root_file, "contains no '", tree_name, "' tree")
+            exit(1)
 
         if maxlen is None:
             self.maxlen = int(self.tree.GetMaximum("n_dau"))

diff --git a/SJ-Keras/Proj-RNN/training.py b/SJ-Keras/Proj-RNN/training.py
@@ -25,20 +25,20 @@
 from pipeline import DataLoader
 # from pipeline import SeqDataLoader as DataLoader
 
-from models import build_a_model
-
 sys.path.append("..")
+from keras4jet.models import build_a_model
 from keras4jet.meters import Meter
-from utils import (
+from keras4jet.utils import (
     get_log_dir,
     get_available_gpus,
     Logger
 )
 
-def main():
+if __name__ == '__main__':
     parser = argparse.ArgumentParser()
 
-    parser.add_argument("--model", type=str, default="model_ak4")
+    parser.add_argument("--model", type=str, default="ak4")
+    parser.add_argument("--directory", type=str, default="../../SJ-JetImage/step5/")
     parser.add_argument("--train_data", type=str, default="dijet")
     parser.add_argument(
 	    '--log_dir', type=str,
@@ -61,13 +61,12 @@ def main():
 
     args = parser.parse_args()
 
-    train_data="../../Data/FastSim_pt_100_500/{}_train.root".format(args.train_data)
-    val_dijet_data="../../Data/FastSim_pt_100_500/dijet_test.root"
-    val_zjet_data="../../Data/FastSim_pt_100_500/zjet_test.root"
+    train_data = args.directory+"/dijet_train.root"
+    val_dijet_data = args.directory+"/dijet_test.root"
+    val_zjet_data = args.directory+"/z_jet_test.root"
 
-    log_dir = get_log_dir(
-        path=args.log_dir.format(name=args.model,
-        creation=True)
+    if '{name}' in args.log_dir: args.log_dir = args.log_dir.format(name=args.model)
+    log_dir = get_log_dir(path=args.log_dir,creation=True)
 
     logger = Logger(log_dir.path, "WRITE")
     logger.get_args(args)
@@ -202,15 +201,7 @@ def main():
     end_time = time.time()
     logger["training_time"] = end_time - start_time
     print("Training is over! :D")
+    filepath = os.path.join(log_dir.saved_models.path, "model_final.h5")
+    model.save(filepath)
     meter.finish()
     logger.finish()
-
-
-
-
-
-
-
-
-if __name__ == "__main__":
- main()
diff --git a/SJ-Keras/keras4jet/__init__.py b/SJ-Keras/keras4jet/__init__.py
@@ -1,9 +1,7 @@
 from __future__ import absolute_import
 
-from keras4jet import losses
-from keras4jet import metrics
-from keras4jet import layers
-from keras4jet import models
+from . import layers
+from . import models
 
-from keras4jet import meters
-from keras4jet import utils
+from . import meters
+from . import utils
diff --git a/SJ-Keras/keras4jet/models/__init__.py b/SJ-Keras/keras4jet/models/__init__.py
@@ -2,3 +2,11 @@
 
 import keras4jet.models.image
 import keras4jet.models.sequential
+
+__all__ = []
+
+def build_a_model(model_name, *args, **kargs):
+    try:
+        return getattr(keras4jet.models.image, model_name).build_a_model(*args, **kargs)
+    except AttributeError:
+        return getattr(keras4jet.models.sequential, model_name).build_a_model(*args, **kargs)
diff --git a/SJ-Keras/keras4jet/models/image/__init__.py b/SJ-Keras/keras4jet/models/image/__init__.py
@@ -1 +1,5 @@
 from __future__ import absolute_import
+
+from . import resnet
+# from . import light_resnet
+from . import cnn
diff --git a/SJ-Keras/keras4jet/models/image/cnn.py b/SJ-Keras/keras4jet/models/image/cnn.py
@@ -0,0 +1,26 @@
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import keras
+import keras.backend as K
+import tensorflow as tf
+
+from keras.layers import Input
+from keras.layers import (
+    Conv2D,
+    GlobalAveragePooling2D,
+    Activation,
+    Dense,
+    Flatten
+)
+from keras.models import Model, Sequential
+
+
+def build_a_model(input_shape, num_classes=1):
+    model = Sequential()
+    model.add(Flatten(input_shape=input_shape))
+    model.add(Dense(100, activation='relu'))
+    model.add(Dense(1, activation='sigmoid'))
+    return model
+
diff --git a/SJ-Keras/keras4jet/models/image/light_resnet.py b/SJ-Keras/keras4jet/models/image/light_resnet.py
@@ -72,7 +72,7 @@ def build_a_model(input_shape, filters_list, num_classes=1):
 
     x = Conv2D(filters=num_classes, kernel_size=1, strides=1)(x)
     logits = GlobalAveragePooling2D()(x) 
-    softmax = Activation('softmax')(logits)
+    sigmoid = Activation('sigmoid')(logits)
 
-    model = Model(inputs=input_image, outputs=softmax)
+    model = Model(inputs=input_image, outputs=sigmoid)
     return model
diff --git a/SJ-Keras/keras4jet/models/image/resnet.py b/SJ-Keras/keras4jet/models/image/resnet.py
@@ -29,9 +29,9 @@ def build_a_model(input_shape, num_classes=1):
 
     x = Conv2D(filters=num_classes, kernel_size=1, strides=1)(x)
     logits = GlobalAveragePooling2D()(x)
-    softmax = Activation('softmax')(logits)
+    sigmoid = Activation('sigmoid')(logits)
 
-    model = Model(inputs=input_image, outputs=softmax)
+    model = Model(inputs=input_image, outputs=sigmoid)
 
     return model
 
diff --git a/SJ-Keras/keras4jet/models/sequential/__init__.py b/SJ-Keras/keras4jet/models/sequential/__init__.py
@@ -6,7 +6,7 @@
 from . import ak4_elu
 from . import ak4_elu_gru
 
-__all__ = ["model_ak4"]
+__all__ = ["ak4"]
 
 def build_a_model(model_name, *args, **kargs):
     return getattr(models, model_name).build_a_model(*args, **kargs)