Cleaned code and revised Readme

README.md

@@ -62,6 +62,7 @@ CNP, LNP and HNP all have a similar encoder-decoder architecture. They have been
 
 
 ## 4. Experiments
+Training can either be conducted in a interactive session (iPython) with arguments set in the section beginning ln 40 (Training parameters). Or by commenting section ln40 and uncommenting section ln 25 (Parse Training parameters) the terminal and it's cmd arguments can be used.
 
 ### 4.1. Regression training
 

examples/deprecated/legacy_train_cnp.py

@@ -1,6 +1,5 @@
 #%%
 import os
-#os.chdir("c:Users/baker/Documents/MLMI4/conditional-neural-processes/")
 import argparse
 from datetime import datetime
 

results/results_graphics.py

@@ -1,6 +1,8 @@
 #%%
-import os
-os.chdir("/Users/baker/Documents/MLMI4/conditional-neural-processes/")
+import os, sys
+pth = os.path.abspath(os.path.join(os.getcwd(), ".."))
+os.chdir(pth) # change to project directory
+sys.path.append(pth)
 from utils.load_model import *
 tf.config.set_visible_devices([], 'GPU') # Disable the GPU if present, we wont need it
 import matplotlib.pyplot as plt
@@ -13,7 +15,6 @@
 pth = f'.data/{args.model}_model_{args.task}_context_{args.num_context}_uniform_sampling_{args.uniform_sampling}/'# + "cp-0030.ckpt"
 model, train_ds, test_ds = load_model_and_dataset(args, pth)
 
-#model.load_weights(pth)
 
 #%%
 
@@ -24,11 +25,6 @@
 #for i, num_context in enumerate([1,10,100,1000]):#([1,10,100,1000]):
 for i, num_context in enumerate([1,10,100,1000]):#([1,10,100,1000]):
 
-    #model.load_weights(f'trained_models/model_{args.task}_context_{num_context}_uniform_sampling_{args.uniform_sampling}/' + "cp-0015.ckpt")
-    #model.load_weights(f'.data/CNP2_model_{args.task}_context_{args.num_context}_uniform_sampling_{args.uniform_sampling}/' + "cp-0010.ckpt")
-
-    #model.load_weights(f'.data/CNP2_model_{args.task}_context_{args.num_context}_uniform_sampling_{args.uniform_sampling}/' + "cp-0010.ckpt")
-
 
     if args.task == 'celeb':
 

results/results_half.py

@@ -1,6 +1,8 @@
 #%%
-import os
-os.chdir("/Users/baker/Documents/MLMI4/conditional-neural-processes/")
+import os, sys
+pth = os.path.abspath(os.path.join(os.getcwd(), ".."))
+os.chdir(pth) # change to project directory
+sys.path.append(pth)
 from utils.load_model import *
 tf.config.set_visible_devices([], 'GPU') # Disable the GPU if present, we wont need it
 from dataloader.load_mnist import half_load_mnist

results/results_latent_prior_vs_posterior.py

@@ -1,6 +1,8 @@
 #%%
-import os
-os.chdir("/Users/baker/Documents/MLMI4/conditional-neural-processes/")
+import os, sys
+pth = os.path.abspath(os.path.join(os.getcwd(), ".."))
+os.chdir(pth) # change to project directory
+sys.path.append(pth)
 from utils.load_model import *
 tf.config.set_visible_devices([], 'GPU') # Disable the GPU if present, we wont need it
 

results/results_regression.py

@@ -1,6 +1,8 @@
 #%%
-import os
-os.chdir("/Users/baker/Documents/MLMI4/conditional-neural-processes/")
+import os, sys
+pth = os.path.abspath(os.path.join(os.getcwd(), ".."))
+os.chdir(pth) # change to project directory
+sys.path.append(pth)
 from utils.load_model import *
 tf.config.set_visible_devices([], 'GPU') # Disable the GPU if present, we wont need it
 import matplotlib.pyplot as plt

results/results_split.py

@@ -1,6 +1,8 @@
 #%%
-import os
-os.chdir("/Users/baker/Documents/MLMI4/conditional-neural-processes/")
+import os, sys
+pth = os.path.abspath(os.path.join(os.getcwd(), ".."))
+os.chdir(pth) # change to project directory
+sys.path.append(pth)
 from utils.load_model import *
 tf.config.set_visible_devices([], 'GPU') # Disable the GPU if present, we wont need it
 from dataloader.load_mnist import split_load_mnist

results/results_varying_kernel.py

@@ -1,6 +1,8 @@
 #%%
-import os
-os.chdir("/Users/baker/Documents/MLMI4/conditional-neural-processes/")
+import os, sys
+pth = os.path.abspath(os.path.join(os.getcwd(), ".."))
+os.chdir(pth) # change to project directory
+sys.path.append(pth)
 from utils.load_model import *
 tf.config.set_visible_devices([], 'GPU') # Disable the GPU if present, we wont need it
 from dataloader.load_regression_data_from_arbitrary_gp_varying_kernel import RegressionDataGeneratorArbitraryGPWithVaryingKernel

train.py

@@ -21,28 +21,29 @@
 tfk = tf.keras
 tfd = tfp.distributions
 
-# Parse arguments
+
+# ================================ Parse Training parameters ===============================================
 # parser = argparse.ArgumentParser()
 # parser.add_argument('-e', '--epochs', type=int, default=120, help='Number of training epochs')
 # parser.add_argument('-b', '--batch', type=int, default=1024, help='Batch size for training')
-# parser.add_argument('-t', '--task', type=str, default='regression', help='Task to perform : (mnist|regression)')
+# parser.add_argument('-t', '--task', type=str, default='regression', help='Task to perform : (mnist|regression|celeb|regression_varying)')
 # parser.add_argument('-c', '--num_context', type=int, default=100)
 # parser.add_argument('-u', '--uniform_sampling', type=bool, default=True)
+# parser.add_argument('-m', '--model', type=bool, default='CNP', help='CNP|LNP|HNP|HNPC')
 # args = parser.parse_args()
 
-
-
+# -------------------------------------------------------------------------------------------------------------------------
 
 
 
 
 # ================================ Training parameters ===============================================
 
 # Regression
-args = argparse.Namespace(epochs=160, batch=1024, task='regression_varying', num_context=25, uniform_sampling=True, model='HNPC')
+args = argparse.Namespace(epochs=160, batch=1024, task='regression', num_context=25, uniform_sampling=True, model='CNP')
 
 # MNIST / Celeb
-#args = argparse.Namespace(epochs=30, batch=256, task='mnist', num_context=100, uniform_sampling=True, model='HNPC')
+#args = argparse.Namespace(epochs=30, batch=256, task='mnist', num_context=100, uniform_sampling=True, model='CNP')
 
 LOG_PRIORS = True
 
@@ -172,8 +173,7 @@ def train_step(model, x, optimizer):
 
 # ============================ Training Loop ===========================================================
 epochs = args.epochs
-#epochs = 160
-for epoch in range(40, epochs + 1):
+for epoch in range(1, epochs + 1):
     with tqdm(total=TRAINING_ITERATIONS, unit='batch') as tepoch:
         tepoch.set_description(f"Epoch {epoch}")