Rangement

higher/smart_aug/datasets.py

@@ -1,3 +1,7 @@
+""" Dataset definition.
+
+    MNIST / CIFAR10
+"""
 import torch
 from torch.utils.data import SubsetRandomSampler
 import torchvision
@@ -28,14 +32,14 @@ transform = torchvision.transforms.Compose([
 #            torchvision.transforms.ToTensor()
 #        ])
 #)
-data_test = torchvision.datasets.MNIST(
-    "./data", train=False, download=True, transform=torchvision.transforms.ToTensor()
-)
+#data_test = torchvision.datasets.MNIST(
+#    "./data", train=False, download=True, transform=torchvision.transforms.ToTensor()
+#)
 
 ### Classic Dataset ###
-data_train = torchvision.datasets.CIFAR10("./data", train=True, download=download_data, transform=transform)
-#data_val = torchvision.datasets.CIFAR10("./data", train=True, download=download_data, transform=transform)
-data_test = torchvision.datasets.CIFAR10("./data", train=False, download=download_data, transform=transform)
+data_train = torchvision.datasets.CIFAR10("../data", train=True, download=download_data, transform=transform)
+#data_val = torchvision.datasets.CIFAR10("../data", train=True, download=download_data, transform=transform)
+data_test = torchvision.datasets.CIFAR10("../data", train=False, download=download_data, transform=transform)
 
 train_subset_indices=range(int(len(data_train)/2))
 val_subset_indices=range(int(len(data_train)/2),len(data_train))

higher/smart_aug/test_brutus.py

@@ -93,7 +93,7 @@ if __name__ == "__main__":
         json.dump(out, f, indent=True)
         print('Log :\"',f.name, '\" saved !')
     '''
-    res_folder="res/brutus-tests2/"
+    res_folder="../res/brutus-tests2/"
     epochs= 150
     inner_its = [1]
     dist_mix = [0.0, 0.5, 0.8, 1.0]
@@ -147,14 +147,14 @@ if __name__ == "__main__":
                             out = {"Accuracy": max([x["acc"] for x in log]), "Time": (np.mean(times),np.std(times), exec_time), 'Optimizer': optim_param, "Device": device_name, "Param_names": aug_model.TF_names(), "Log": log}
                             print(str(aug_model),": acc", out["Accuracy"], "in:", out["Time"][0], "+/-", out["Time"][1])
                             filename = "{}-{} epochs (dataug:{})- {} in_it-{}".format(str(aug_model),epochs,dataug_epoch_start,n_inner_iter, run)
-                            with open("res/log/%s.json" % filename, "w+") as f:
+                            with open("../res/log/%s.json" % filename, "w+") as f:
                                 try:
                                     json.dump(out, f, indent=True)
                                     print('Log :\"',f.name, '\" saved !')
                                 except:
                                     print("Failed to save logs :",f.name)
                             try:
-                                plot_resV2(log, fig_name="res/"+filename, param_names=aug_model.TF_names())
+                                plot_resV2(log, fig_name="../res/"+filename, param_names=aug_model.TF_names())
                             except:
                                 print("Failed to plot res")
 

higher/smart_aug/test_dataug.py

@@ -1,8 +1,13 @@
+""" Script to run experiment on smart augmentation.
+
+"""
+
 from model import *
 from dataug import *
 #from utils import *
 from train_utils import *
 
+# Use available TF (see transformations.py)
 tf_names = [
     ## Geometric TF ##
     'Identity',
@@ -34,6 +39,8 @@ tf_names = [
     #'=Posterize',
     #'=Solarize',
 
+    ## Bad Tranformations ##
+    # Bad Geometric TF #
     #'BShearX',
     #'BShearY',
     #'BTranslateX-', 
@@ -46,12 +53,14 @@ tf_names = [
 
     #'Random',
     #'RandBlend'
+
     #Non fonctionnel
     #'Auto_Contrast', #Pas opti pour des batch (Super lent)
     #'Equalize',
 ]
 
-device = torch.device('cuda')
+
+device = torch.device('cuda') #Select device to use
 
 if device == torch.device('cpu'):
     device_name = 'CPU'
@@ -61,13 +70,15 @@ else:
 ##########################################
 if __name__ == "__main__":
 
+    #Task to perform
     tasks={
         #'classic',
-        #'aug_dataset',
+        #'aug_dataset', #Moved to old code
         'aug_model'
     }
+    #Parameters
     n_inner_iter = 1
-    epochs = 150
+    epochs = 200
     dataug_epoch_start=0
     optim_param={
         'Meta':{
@@ -81,6 +92,7 @@ if __name__ == "__main__":
         }
     }
 
+    #Models
     model = LeNet(3,10)
     #model = ResNet(num_classes=10)
     #Lents
@@ -103,17 +115,18 @@ if __name__ == "__main__":
         out = {"Accuracy": max([x["acc"] for x in log]), "Time": (np.mean(times),np.std(times), exec_time), 'Optimizer': optim_param['Inner'], "Device": device_name, "Log": log}
         print(str(model),": acc", out["Accuracy"], "in:", out["Time"][0], "+/-", out["Time"][1])
         filename = "{}-{} epochs".format(str(model),epochs)
-        with open("res/log/%s.json" % filename, "w+") as f:
+        with open("../res/log/%s.json" % filename, "w+") as f:
             json.dump(out, f, indent=True)
             print('Log :\"',f.name, '\" saved !')
 
-        plot_res(log, fig_name="res/"+filename)
+        plot_res(log, fig_name="../res/"+filename)
 
         print('Execution Time : %.00f '%(exec_time))
         print('-'*9)
     
 
     #### Augmented Dataset ####
+    '''
     if 'aug_dataset' in tasks:
 
         t0 = time.process_time()
@@ -162,7 +175,7 @@ if __name__ == "__main__":
 
         print('Execution Time : %.00f '%(exec_time))
         print('-'*9)
-    
+    '''
 
     #### Augmented Model ####
     if 'aug_model' in tasks:
@@ -170,7 +183,7 @@ if __name__ == "__main__":
 
         tf_dict = {k: TF.TF_dict[k] for k in tf_names}
         model = Higher_model(model) #run_dist_dataugV3
-        aug_model = Augmented_model(Data_augV5(TF_dict=tf_dict, N_TF=3, mix_dist=0.8, fixed_prob=False, fixed_mag=False, shared_mag=False), model).to(device)
+        aug_model = Augmented_model(Data_augV7(TF_dict=tf_dict, N_TF=3, mix_dist=0.8, fixed_prob=False, fixed_mag=False, shared_mag=False), model).to(device)
         #aug_model = Augmented_model(RandAug(TF_dict=tf_dict, N_TF=2), model).to(device)
 
         print("{} on {} for {} epochs - {} inner_it".format(str(aug_model), device_name, epochs, n_inner_iter))

higher/smart_aug/train_utils.py

@@ -263,8 +263,8 @@ def run_dist_dataugV3(model, opt_param, epochs=1, inner_it=1, dataug_epoch_start
 
         if (save_sample_freq and epoch%save_sample_freq==0): #Data sample saving
                 try:
-                    viz_sample_data(imgs=xs, labels=ys, fig_name='samples/data_sample_epoch{}_noTF'.format(epoch))
-                    viz_sample_data(imgs=model['data_aug'](xs), labels=ys, fig_name='samples/data_sample_epoch{}'.format(epoch))
+                    viz_sample_data(imgs=xs, labels=ys, fig_name='../samples/data_sample_epoch{}_noTF'.format(epoch))
+                    viz_sample_data(imgs=model['data_aug'](xs), labels=ys, fig_name='../samples/data_sample_epoch{}'.format(epoch))
                 except:
                     print("Couldn't save samples epoch"+epoch)
                     pass
@@ -327,8 +327,8 @@ def run_dist_dataugV3(model, opt_param, epochs=1, inner_it=1, dataug_epoch_start
 
     #Data sample saving
     try:
-        viz_sample_data(imgs=xs, labels=ys, fig_name='samples/data_sample_epoch{}_noTF'.format(epoch))
-        viz_sample_data(imgs=model['data_aug'](xs), labels=ys, fig_name='samples/data_sample_epoch{}'.format(epoch))
+        viz_sample_data(imgs=xs, labels=ys, fig_name='../samples/data_sample_epoch{}_noTF'.format(epoch))
+        viz_sample_data(imgs=model['data_aug'](xs), labels=ys, fig_name='../samples/data_sample_epoch{}'.format(epoch))
     except:
         print("Couldn't save finals samples")
         pass

higher/smart_aug/utils.py

@@ -1,3 +1,6 @@
+""" Utilties function.
+
+"""
 import numpy as np
 import json, math, time, os
 import matplotlib.pyplot as plt
@@ -12,12 +15,24 @@ import torch.nn.functional as F
 import time
 
 def print_graph(PyTorch_obj, fig_name='graph'):
-    graph=make_dot(PyTorch_obj) #Loss give the whole graph
+    """Save the computational graph.
+
+        Args:
+            PyTorch_obj (Tensor): End of the graph. Commonly, the loss tensor to get the whole graph.
+            fig_name (string): Relative path where to save the graph. (default: graph)
+    """
+    graph=make_dot(PyTorch_obj)
     graph.format = 'pdf' #https://graphviz.readthedocs.io/en/stable/manual.html#formats
     graph.render(fig_name)
 
 def plot_resV2(log, fig_name='res', param_names=None):
+    """Save a visual graph of the logs.
 
+        Args:
+            log (dict): Logs of the training generated by most of train_utils.
+            fig_name (string): Relative path where to save the graph. (default: res)
+            param_names (list): Labels for the parameters. (default: None)
+    """
     epochs = [x["epoch"] for x in log]
 
     fig, ax = plt.subplots(nrows=2, ncols=3, figsize=(30, 15))
@@ -63,7 +78,12 @@ def plot_resV2(log, fig_name='res', param_names=None):
     plt.close()
 
 def plot_compare(filenames, fig_name='res'):
+    """Save a visual graph comparing trainings stats.
 
+        Args:
+            filenames (list[Strings]): Relative paths to the logs (JSON files).
+            fig_name (string): Relative path where to save the graph. (default: res)
+    """
     all_data=[]
     legend=""
     for idx, file in enumerate(filenames):
@@ -102,21 +122,15 @@ def plot_compare(filenames, fig_name='res'):
     plt.savefig(fig_name, bbox_inches='tight')
     plt.close()
 
-def plot_TF_res(log, tf_names, fig_name='res'):
-
-    mean = np.mean([x["param"] for x in log], axis=0)
-    std = np.std([x["param"] for x in log], axis=0)
-
-    fig, ax = plt.subplots(1, 1, figsize=(30, 8), sharey=True)
-    ax.bar(tf_names, mean, yerr=std)
-    #ax.bar(tf_names, log[-1]["param"])
-
-    fig_name = fig_name.replace('.',',')
-    plt.savefig(fig_name, bbox_inches='tight')
-    plt.close()
-
 def viz_sample_data(imgs, labels, fig_name='data_sample', weight_labels=None):
+    """Save data samples.
 
+        Args:
+            imgs (Tensor): Batch of image to sample from. Intended to contain at least 25 images.
+            labels (Tensor): Labels of the images.
+            fig_name (string): Relative path where to save the graph. (default: data_sample)
+            weight_labels (Tensor): Weights associated to each labels. (default: None)
+    """
     sample = imgs[0:25,].permute(0, 2, 3, 1).squeeze().cpu()
 
     plt.figure(figsize=(10,10))
@@ -135,7 +149,11 @@ def viz_sample_data(imgs, labels, fig_name='data_sample', weight_labels=None):
     plt.close('all')
 
 def print_torch_mem(add_info=''):
+    """Print informations on PyTorch memory usage.
 
+        Args:
+            add_info (string): Prefix added before the print. (default: None)
+    """
     nb=0
     max_size=0
     for obj in gc.get_objects():
@@ -165,6 +183,7 @@ def print_torch_mem(add_info=''):
         torch.cuda.max_memory_cached()/ mega_bytes)
     print(string)
 
+'''
 def plot_TF_influence(log, fig_name='TF_influence', param_names=None):
     proba=[[x["param"][idx]['p'] for x in log] for idx, _ in enumerate(log[0]["param"])]
     mag=[[x["param"][idx]['m'] for x in log] for idx, _ in enumerate(log[0]["param"])]
@@ -179,21 +198,24 @@ def plot_TF_influence(log, fig_name='TF_influence', param_names=None):
     fig_name = fig_name.replace('.',',')
     plt.savefig(fig_name, bbox_inches='tight')
     plt.close()
+'''
 
-### https://github.com/facebookresearch/higher/issues/18 ####
 from torch._six import inf
 def clip_norm(tensors, max_norm, norm_type=2):
-    r"""Clips norm of passed tensors.
-    The norm is computed over all tensors together, as if they were
-    concatenated into a single vector. Clipped tensors are returned.
-    Arguments:
-        tensors (Iterable[Tensor]): an iterable of Tensors or a
-            single Tensor to be normalized.
-        max_norm (float or int): max norm of the gradients
-        norm_type (float or int): type of the used p-norm. Can be ``'inf'`` for
-            infinity norm.
-    Returns:
-      Clipped (List[Tensor]) tensors.
+    """Clips norm of passed tensors.
+        The norm is computed over all tensors together, as if they were
+        concatenated into a single vector. Clipped tensors are returned.
+        
+        See: https://github.com/facebookresearch/higher/issues/18
+
+        Args:
+            tensors (Iterable[Tensor]): an iterable of Tensors or a
+                single Tensor to be normalized.
+            max_norm (float or int): max norm of the gradients
+            norm_type (float or int): type of the used p-norm. Can be ``'inf'`` for
+                infinity norm.
+        Returns:
+          Clipped (List[Tensor]) tensors.
     """
     if isinstance(tensors, torch.Tensor):
         tensors = [tensors]