diff --git a/.DS_Store b/.DS_Store
index 54e3650..0f73a22 100644
Binary files a/.DS_Store and b/.DS_Store differ
diff --git a/code/.DS_Store b/code/.DS_Store
new file mode 100644
index 0000000..450e09f
Binary files /dev/null and b/code/.DS_Store differ
diff --git a/code/best_mem.pickle b/code/best_mem.pickle
deleted file mode 100755
index 77d51ce..0000000
Binary files a/code/best_mem.pickle and /dev/null differ
diff --git a/code/open_mem.py b/code/open_mem.py
deleted file mode 100644
index 7686505..0000000
--- a/code/open_mem.py
+++ /dev/null
@@ -1,121 +0,0 @@
-#%%
-import pickle
-import os
-import torch
-import clip
-import time
-import pickle
-import random
-import argparse
-import torch.nn as nn
-import torch.optim as optim
-from PIL import Image
-
-from torch.utils.data import DataLoader
-
-from config import *
-from dataset import *
-from models import *
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
-path = 'best_mem.pickle'
-
-with open(path, 'rb') as f:
-    memory = pickle.load(f)
-#%%
-in_path = '/Users/filippomerlo/Desktop/Datasets/SOLA'
-source = 'train'
-memory = memory
-in_base = bn_train
-types = tyimgs
-dic = dic_train_logical
-vocab = all_vocabs
-# %%	
-def my_clip_evaluation(in_path, source, memory, in_base, types, dic, vocab):
-
-	with torch.no_grad():
-		# get vocab dictionary
-		if source == 'train':
-			dic = dic_test
-		else:
-			dic = dic_train
-
-		# get dataset
-		clip_model, clip_preprocess = clip.load("ViT-B/32", device=device)
-		dt = MyDataset(in_path, source, in_base, types, dic, vocab,
-					clip_preprocessor=clip_preprocess)
-		data_loader = DataLoader(dt, batch_size=132, shuffle=True)
-
-		top3 = 0
-		top3_color = 0
-		top3_material = 0
-		top3_shape = 0
-		tot_num = 0
-		i = 0
-		#for base_is, images in data_loader: # labels (one hot), images (clip embs)
-		base_is, images = next(iter(data_loader))
-		# Prepare the inputs
-		images = images.to(device)
-		ans = []
-		rel = []
-		batch_size_i = len(base_is)
-
-		# go through memory
-		for label in vocab: # select a label es 'red'
-			if label not in memory.keys():
-				ans.append(torch.full((batch_size_i, 1), 1000.0).squeeze(1))
-				continue
-
-			# load model
-			model = CLIP_AE_Encode(hidden_dim_clip, latent_dim, isAE=False)
-			model.load_state_dict(memory[label]['model']) # load weights corresponding to red
-			model.to(device)
-			model.eval() # freeze
-
-			# load centroid
-			centroid_i = memory[label]['centroid'].to(device)
-			centroid_i = centroid_i.repeat(batch_size_i, 1)
-
-			# compute stats
-			z = model(clip_model, images).squeeze(0)
-			disi = ((z - centroid_i)**2).mean(dim=1)
-			ans.append(disi.detach().to('cpu'))
-
-		# get top3 indices
-		ans = torch.stack(ans, dim=1)
-		values, indices = ans.topk(3, largest=False)
-		_, indices_lb = base_is.topk(3)
-		# base_is = [00001000000010000001]
-		# indices_lb [5,12,19]
-		indices_lb, _ = torch.sort(indices_lb)
-
-		# calculate stats
-		tot_num += len(indices)
-		for bi in range(len(indices)):
-			ci = 0
-			mi = 0
-			si = 0
-			print('***',indices[bi],'***')
-			if indices_lb[bi][0] in indices[bi]:
-				print(indices_lb[bi][0])
-				ci = 1
-			if indices_lb[bi][1] in indices[bi]:
-				print(indices_lb[bi][1])
-				mi = 1
-			if indices_lb[bi][2] in indices[bi]:
-				print(indices_lb[bi][2])
-				si = 1
-
-			top3_color += ci
-			top3_material += mi
-			top3_shape += si
-			if (ci == 1) and (mi == 1) and (si == 1):
-				top3 += 1
-
-		print(tot_num, top3_color/tot_num, top3_material/tot_num,
-				top3_shape/tot_num, top3/tot_num)
-	return top3/tot_num
-
-n = my_clip_evaluation(in_path, source, memory, in_base, types, dic, vocab)
-print(n)
\ No newline at end of file
diff --git a/code/try.py b/code/try.py
deleted file mode 100644
index 1f99422..0000000
--- a/code/try.py
+++ /dev/null
@@ -1,40 +0,0 @@
-#%%
-from dataset import *
-from config import *
-
-from torch.utils.data import DataLoader
-from pprint import pprint
-
-import matplotlib.pyplot as plt
-
-in_path = '/Users/filippomerlo/Desktop/Datasets/SOLA'
-source = 'train'
-in_base = bn_train
-types = tyimgs
-dic = dic_train_logical
-vocab = all_vocabs
-
-dt = MyDataset(in_path, source, in_base, types, dic, vocab)
-data_loader = DataLoader(dt, batch_size=132, shuffle=True)
-
-#%%
-train_labels, train_features = next(iter(data_loader))
-print(f"Feature batch shape: {train_features.size()}")
-print(f"Labels batch shape: {train_labels.size()}")
-pprint(train_features)
-
-#%% Class Methods
-#dt.__len__()
-a,b = dt.__getitem__(1240)
-print(a)
-#%% NEW Functions
-# - With attributes pairing between positive and negative samples 
-
-attr = 'color'
-lesson = 'red'
-names_sim, images_sim, names_dif, images_dif = dt.get_paired_batches(attr,lesson)
-
-for i,n in enumerate(names_sim):
-    print('**********',i,'**********')
-    print(n,'\n',names_dif[i])
-