-
Notifications
You must be signed in to change notification settings - Fork 0
/
Agent.py
643 lines (527 loc) · 26.6 KB
/
Agent.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
'''DOC-STRING
->
DEVELOPER-NOTES
-> This code is the developed version of running.py with added cross-attention layer in the NN.
-> No improvments after this model
->it has a change in command-None code the latest version from 3 and 2
'''
import pexpect
import regex
import os
import nltk
import numpy as np
from datetime import datetime
from nltk.corpus import stopwords
#from nltk.corpus import wordnet as wn
from nltk.corpus import brown
from gensim.models import Word2Vec
from sentence_transformers import SentenceTransformer
import spacy
from transformers import BertTokenizer, TFBertModel,TFDistilBertModel,DistilBertTokenizer
import matplotlib.pyplot as plt
import tensorflow as tf
from libraries.Experience_replay import experience_replay
from libraries.graph import graph_node
import libraries.function_dup as fns
from keras.losses import Loss
from striprtf.striprtf import rtf_to_text
import pandas as pd
from libraries.cross_attention import Cross_Attention_Layer
nlp = spacy.load('en_core_web_sm')
commands_dict = {}
#Reading the file to get commands
f1 = open("commands.rtf","r")
commands = rtf_to_text(f1.read())
commands = commands.split(sep = "\n")
#commands = f1.readlines()
#print(len(commands))
# Loading the game
game = pexpect.spawn("frotz game/zork1.z5")
#ALL Regex patterns for retrieving the texts
pattern = r"(\r.*(\[){1}m)"
pattern2 = r"(\r.*[0-9]d)"
#pattern3 = r"\x1b.*H"
initial = r"\x1b.*(\[){1}m"
#pos = "[A-Z][a-z].*[a-z]{2,}"
only_pat = regex.compile(r"((\x9B|\x1B\[)[0-?]*[ -\/]*[@-~])|(\x1b\(B)")
#pos_pattern = regex.compile(pos)
initial_pattern = regex.compile(initial)
r = regex.compile(pattern)
r2 = regex.compile(pattern2)
#r3 = regex.compile(pattern3)
#Getting the attributes for the syntax specified
def get_command_attr(command_syntax,items,objects):
parts = command_syntax.split()
if "[item]" in parts:
if "[object]" in parts:
df = fns.get_simiarity_score(command_syntax,type=None,items = items,objects = objects)
#print("\nDataFrame : ",df)
#getting the command with the highest similarity
try:
command = df[df["similarity"] == df["similarity"].min()]["command"].values.item()
except:
command = None
#print("\nSelected command : ", type(command),"\n")
return command,df
else:
df = fns.get_simiarity_score(command_syntax,type=None,items = items,objects = None)
#print("\nDataFrame : ",df)
#getting the command with the highest similarity
try:
command = df[df["similarity"] == df["similarity"].min()]["command"].values.item()
except:
command = None
#print("\nSelected command : ", command,"\n")
return command,df
elif "[object]" in parts:
if parts.count("[object]") == 2:
df = fns.get_simiarity_score(command_syntax,type="double",items = None,objects = objects)
#print("\nDataFrame : ",df)
#getting the command with the highest similarity
try:
command = df[df["similarity"] == df["similarity"].min()]["command"].values.item()
except:
command = None
#print("\nSelected command : ", command,"\n")
return command,df
else:
df = fns.get_simiarity_score(command_syntax,type="simple",items = None,objects = objects)
#print("\nDataFrame : ",df)
#getting the command with the highest similarity
try:
command = df[df["similarity"] == df["similarity"].min()]["command"].values.item()
except:
command = None
#print("\nSelected command : ", type(command),"\n")
return command,df
else:
df = pd.DataFrame({"similarity" : 1, "command" : command_syntax},index=[0])
return command_syntax,df
#Custom loss function
def logloss(y_expec, y_pred):
y_expec = tf.clip_by_value(y_expec,clip_value_min=0.001,clip_value_max=1)
log_prob = -1*tf.math.log(y_pred)*y_expec
return log_prob
#get X anf Y traninig data
def get_training(batches,discount_factor = 0.6):
X = np.zeros(shape=(batches[0].shape[0],150,768))
X_action = np.zeros(shape=(batches[0].shape[0],12,768))
Y_statevalue = np.zeros(shape=(batches[0].shape[0],1,1))
Y_qvalue = np.zeros(shape = (batches[0].shape[0],1,75))
dis_reward = 0
for i in range(batches[0].shape[0]-1,-1,-1):
#Setting the X values for training
X[i,:,:] = batches[0][i,:,:]
X_action[i,:,:] = batches[3][i,:,:]
#Calculating Cumilative reward with given discount factor
dis_reward = batches[2][i] + discount_factor*dis_reward
Y_statevalue[i,0,0] = dis_reward
Y_qvalue[i,0,:].fill(dis_reward - batches[-1][i])
'''for iter in batches:
dis_reward = 0
#print("Each batch shape : ",iter[0].shape[0])
for j in range(iter[0].shape[0]):
#print("iter shape : ",iter[0].shape,end="\n")
#Calculating Cumilative reward with given discount factor
dis_reward = iter[2][j] + discount_factor*dis_reward
#Setting the X values for training
X[j,:,:] = iter[0][j,:,:]
X_action[j,:,:] = iter[3][j,:,:]
#The result of the value function is the cumilative reward Gi
Y_statevalue[j,0] = dis_reward
#Advantage value and training result for Q value
delta = dis_reward - iter[-1][j]
Y_qvalue[j,:] = delta
#np.log(iter[1][j])*delta'''
return X,X_action,Y_qvalue,Y_statevalue
#Function to get the first output
def get_initial_output():
state_info = {}
#output for the first iterstion is treated differently as it has unwanted information
game.expect(">")
output = game.before.decode("utf-8")
position1 = "West of House"
get_rid = "ZORK I: The Great Underground Empire Copyright (c) 1981, 1982, 1983 Infocom, Inc. All rights reserved. ZORK is a registered trademark of Infocom, Inc. Revision 88 / Serial number 840726 West of House "
output = regex.sub(initial_pattern,"",output)
output_first = output.split(" ")
for i,j in zip(output_first,range(len(output_first))):
output_first[j] = regex.sub(r2," ",i)
output_first = " ".join(output_first)
output_first = output_first.replace(get_rid,"")
state_info["position"] = position1
state_info["description"] = output_first
#print(state_info)
return state_info
#Function to get the state info for the given command
def get_state_variables(command,prev_pos):
state_info = {}
# if the game is restarted then the intial output function is called
if command == "restart":
global game
#game.timeout = 1
#game.expect([">",pexpect.TIMEOUT])
#game.sendline("restart" + "\n")
game.terminate(force=True)
game = pexpect.spawn("frotz game/zork1.z5")
return None,None#get_initial_output()
#Sending the first input and getting the output to and from the game
game.timeout = 1
game.expect([">",pexpect.TIMEOUT])
game.sendline(command + "\n")
game.expect([">",pexpect.TIMEOUT])
output1 = game.before.decode("utf-8")
#print("OUTPUT RECIEVED : ",output1.encode("utf-8"))
#Getting the position from the output
lines = only_pat.sub(" ",output1).split(sep="\r")
#print("LINES : ",lines)
score=0
#If the word entered is not recognised then there is a spcial if statement
if 'know the word' in lines[1].strip() or not(lines[1][-1].isdigit()):
output1 = lines[1].strip()
state_info["position"] = prev_pos
#If the output is time passes
elif lines[2].strip() == "Time passes...":
state_info["position"] = prev_pos
position = prev_pos
lines = "Time passes..."
#Getting the score
score = lines[1][-3] if lines[1][-3].isdigit() else 0
#elif statement to get the name of the place otherwise
elif (len(lines[2].split())>3) or lines[2].strip() in ("Taken.","Aaaarrrrgggghhhh!") or lines[2].endswith((".","!","?")):
#print("entered here")
state_info["position"] = prev_pos
position = prev_pos
output1 = " ".join(lines[2:])
#Getting the score
score = lines[1][-3] if lines[1][-3].isdigit() else 0
#When entering new place name is taken
else:
position = lines[2]
state_info["position"] = position.strip()
output1 = " ".join(lines[3:])
#Getting the score
score = lines[1][-3] if lines[1][-3].isdigit() else 0
'''if (len(lines[2].split())>3): position = '0'
else: position = lines[2]
if position == '0':
state_info["position"] = prev_pos
position = prev_pos
output1 = " ".join(lines[2:])
else:
state_info["position"] = position.strip()
output1 = " ".join(lines[3:])'''
#moves = lines[1]
state_info["description"] = output1.strip()
#if position in lines[1]:
# moves = moves[-1]
#print(state_info)
return state_info,score
#Function to add the missing values in the input
def add_empty(container,type="description"):
if type == "description":
zeros = tf.zeros(shape=(1,150-container.shape[1],container.shape[-1]))
container = tf.concat([container,zeros],axis=1)
#print("Container shape : ",container.shape,end="\n")
return container
else:
zeros = tf.zeros(shape=(1,12-container.shape[1],container.shape[-1]))
container = tf.concat([container,zeros],axis=1)
#print("Container shape : ",container.shape,end="\n")
return container
#Function to get the command embddings - Not sure whether to be used
def get_command_embedding(command):
if command is None:
return tf.zeros(shape=(1,12,768))
tokeniser = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
model = TFDistilBertModel.from_pretrained("distilbert-base-uncased")
words = tokeniser(command,return_tensors = "tf")
output = model(words)
emb = output[0][:,1:-1,:]
emb = add_empty(emb,type="command")
return emb
#Fuction to create the keras lstm model (triggered only once) - ADD NORMALISATION LAYER IF NECESSARY
def create_tf_model():
#Input layer
input = tf.keras.layers.Input(shape=(150,768))
input_action = tf.keras.layers.Input(shape=(12,768))
#Cross-Attention Layer to get sentence sense
attn = Cross_Attention_Layer(768,768,128,64)([input,input_action])#(tf.concat([input,input_action],axis = 0))
#print(attn.shape)
#LSTM Layer with 100 cells
lstm = tf.keras.layers.LSTM(100,input_shape = (150,768),recurrent_activation = "sigmoid")(tf.expand_dims(attn,axis=0))
print("\ncrossed LSTM : \n",lstm.shape,end="\n")
norm = tf.keras.layers.Normalization(axis=-1)(lstm)
#ANN for Q-value for actions
norm = tf.keras.layers.Dropout(0.01)(norm)
dense1_action = tf.keras.layers.Dense(512,activation='elu')(norm)
dense2_action = tf.keras.layers.Dense(128,activation='elu')(dense1_action)
dense3_action = tf.keras.layers.Dense(75,activation='softmax',name = "q_value")(dense2_action)
#ANN for value funtion for state
dense1_state = tf.keras.layers.Dense(64,activation='elu')(norm)
dense2_state = tf.keras.layers.Dense(16,activation='sigmoid')(dense1_state)
dense3_state = tf.keras.layers.Dense(1,activation='relu',name = "state_value")(dense2_state)
print("\npassed the model creation\n")
#Model compilation
model = tf.keras.Model(inputs = [input,input_action],outputs = [dense3_action,dense3_state])
model.compile(optimizer = "Adam",loss={"q_value" : logloss, "state_value" : "mean_squared_error"}, metrics="accuracy")
return model
# Function to run the bert and pass through lstm
def NLP_model(descp,action_taken,model):
#Removing Stop words
# Initializing bert tokenizer and bert pre-trained model
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
bert = TFBertModel.from_pretrained("bert-base-uncased")
#For description
#Tokenising the sentence and turning it into ids with masking
words = tokenizer.encode(descp)
#stp_wrds = set(stopwords.words("english")) - set(("not","no","never","hasn't","can't","won't","hadn't","couldn't","wouldn't","shouldn't","haven't","mustn't","isn't"))
#words = [i for i in words if i not in stp_wrds]
#ids = tokenizer.convert_tokens_to_ids(words)
#ids = tf.expand_dims(ids, 0)
words = tf.expand_dims(words, 0)
mask = tf.ones_like(words)
#Getting the output from the bert model
outputs = bert(words, attention_mask=mask)
final = outputs[0]
#Zeros added to match the dimension of the lstm input
final = add_empty(final,type="description")
if action_taken is not None:
#For command
#tokenising the sentence and turining into ids
'''command_words = tokenizer.encode(descp)
#ids = tokenizer.convert_tokens_to_ids(words)
#ids = tf.expand_dims(ids, 0)
command_words = tf.expand_dims(command_words, 0)
mask = tf.ones_like(command_words)
#Getting the output from the bert model
command_output = bert(command_words, attention_mask=mask)
command_emb = command_output[0]
#Zeros added to match the dimension of the lstm input
print(command_emb.shape,"\n\n")
command_emb = add_empty(command_emb,type="command")'''
action_taken = get_command_embedding(action_taken)#command_emb
else:
action_taken = tf.zeros(shape=(1,12,768))
#Check GPU and CPU
'''devices = tf.config.list_physical_devices("GPU")
for i in devices:
print("\ndevice type : ",i.device_type)
print("\nfreq : ",tf.config.experimental.get_device_details(i))'''
#getting states from the model with the data
#print(final.shape)
action_prob = model.predict([final,action_taken])
print(action_prob[1].shape)
return [final,action_prob]
#Function to update the minimap
def record_minimap(dest_name,head_node=None,direction=None):
if head_node == None:
head_node = graph_node(dest_name)
return head_node
else:
head_node.set_n_node(dest_name,direction)
return head_node.n_nodes[-1][0]
#Main function that controls the flow of the program
def propogate(**kwargs):
episode_count = 0
#Object and tuple creation for model and experience replay
model = create_tf_model()
exp = experience_replay()
score_dict = {"West of House" : {}}
#NLP_model(info["description"],model=model)
epsilon_value = 0.8
if len(kwargs)!=0:
model = kwargs["model"]
exp = kwargs["exp"]
score_dict = kwargs["score_dict"]
episode_count = kwargs["episode"]+1
#Episode loop each episode consists of 100 iterations
while(episode_count<10):
i=1
#Getting the initial state information and minimap
info = get_initial_output()
start_location = record_minimap(info["position"])
current_location = start_location
object_dict = {}
#required items, objects and environment related details
items = []
visited = ()
area_look = {}
area_look[info["position"]] = info["description"]
action_taken = None
prev_action_taken = None
reward_object = fns.Reward(0)
while(i<=1000):
print("\nIteration : ",i,"\n")
#print("\nState_Info : ",info,end="\n")
#Passing it through the nlp model
[state,[prob,value]] = NLP_model(info["description"],action_taken,model=model)
#print("VALUE FUNCTION RESULT : ",value)
#print("Prob shape of output : ",prob.shape)
#Getting the index of the action to be performed using epsilon greedy
index = fns.epsilon_greedy(prob,epsilon=epsilon_value)
#Getting the valid action and if inventory is emoty all the commands with [item] tag is removed
if items == []:
prob_copy = prob
for index in range(len(commands)):
if "[item]" in commands[index]:
#print(commands[index])
prob = np.delete(prob,index,axis=1)
#print(prob)
index = fns.epsilon_greedy(prob,epsilon=epsilon_value,empty_inventory=1)
action_syntax = commands[index]
objects = fns.get_objects(info["description"])
if info["position"] in object_dict:
if object_dict[info["position"]] == 1:
score_dict[info["position"]] = {}
object_dict[info["position"]] = 0
#If actions have not been taken in that position
if info["position"] not in score_dict:
action_taken,df = get_command_attr(action_syntax,items,objects+items)
print("at first : ",action_taken)
#Getting the dataframe with some command to execute while the preferences are none
if action_taken is None:
num = 2
while len(df) == 0 or (action_taken is None):
if num>len(prob[0]):
return {"model":model,"exp":exp,"score_dict":score_dict,"episode":episode_count}
index = fns.epsilon_greedy(prob,epsilon=epsilon_value,empty_inventory=num)
action_syntax = commands[index]
action_taken,df = get_command_attr(action_syntax,items,objects+items)
print("after : ",action_taken,df)
num+=1
#The action syntax is stored for that position
if len(df) == 0:
break
score_dict[info["position"]] = {action_syntax : df}
print("ACTION_DICT : ",score_dict[info["position"]].keys(),action_taken)
#If actions have been already taken in that state
elif info["position"] in score_dict:
#print("ACTION : ",action_syntax ,"DICT : ",score_dict[info["position"]].keys())
#if the action has already been taken before it takes the old dataframe
if action_syntax in score_dict[info["position"]]:
df = score_dict[info["position"]][action_syntax]
num = 2
#If action_syntax in score_dict but the df is empty - meaning the pbjects that existed earlier did not provide positive results
while len(df) == 0:
if num>len(prob[0]):
return {"model":model,"exp":exp,"score_dict":score_dict,"episode":episode_count}
index = fns.epsilon_greedy(prob,epsilon=1,empty_inventory=num)
action_syntax = commands[index]
if action_syntax not in score_dict[info["position"]]:
action_taken,df = get_command_attr(action_syntax,items,objects+items)
else:
df = score_dict[info["position"]][action_syntax]
if action_syntax not in score_dict[info["position"]]:
score_dict[info["position"]][action_syntax] = df
else:
action_taken = df[df["similarity"] == df["similarity"].min()]["command"].values.item()
#If the action is new the attributes are decided
else:
action_taken,df = get_command_attr(action_syntax,items,objects+items)
#Getting the dataframe with some command to execute while the preferences are none
if action_taken is None:
num=2
while len(df) == 0 or action_taken == None:
if num>len(prob[0]):
return {"model":model,"exp":exp,"score_dict":score_dict,"episode":episode_count}
index = fns.epsilon_greedy(prob,epsilon=epsilon_value,empty_inventory=num)
action_syntax = commands[index]
#If the command is already present in the dictionary then the record is considered
if action_syntax in score_dict[info["position"]]:
df = score_dict[info["position"]][action_syntax]
try:
action_taken = df[df["similarity"] == df["similarity"].min()]["command"].values.item()
except:
action_syntax = None
else:
action_taken,df = get_command_attr(action_syntax,items,objects+items)
num+=1
if action_syntax not in score_dict[info["position"]]:
score_dict[info["position"]][action_syntax] = df
else:
score_dict[info["position"]][action_syntax] = df
print("\nAction Taken : ",action_taken)
#Recording the minimap
if info["position"] in visited:
pass
else:
visited+=(info["position"],)
current_location = record_minimap(info["position"],current_location,action_taken)
#Adding items to items
if action_syntax.startswith("take"):
items.append(action_taken.split()[-1])
#print("\nItems : ",items,"\n")
elif action_syntax.startswith("drop"):
items.remove(action_taken.split()[-1])
#print("\nItems : ",items,"\n")
#Executing the action taken and resukt is stored for the next state
next_info,score = get_state_variables(action_taken,current_location.name)
next_info["description"] = next_info["description"].replace('"',"").replace("can't","cannot")
print("RESPONSE : ",next_info,end="\n")
#Getting the reward for the action
if next_info["description"] == "":
#reward = reward_object.reward_function({"position" : next_info["position"], "description" : area_look[next_info["position"]]},score)
reward = 0
else:
reward = reward_object.reward_function(next_info,int(score))
if reward<0:
print("NEGATIVE")
else:
print("POSITIVE")
#changing the value for each action based on the reward and storing it in a dictionary
table = score_dict[info["position"]][action_syntax]
#print("dataframe to be considered : \n",table,"action_taken : ",action_taken)
sim_score = table[table["command"] == action_taken].iloc[-1,0]
#print("Score of the chosen action : ",sim_score)
if reward<=0:
table = table[table["similarity"] != sim_score]
score_dict[info["position"]][action_syntax] = table
info = next_info
#Adding the sequence to the experience replay
exp.add_sars(np.reshape(state,(150,768)),prob[0,index],reward,get_command_embedding(prev_action_taken),False,value)
prev_action_taken = action_taken
#To concatenate the descriptions for the next iteration
if info["position"] in area_look:
#If the reward is negative then the original state description is taken
if reward<0:
info["description"] = area_look[info["position"]]
#Else if reward is positive the description is concatenated to the original one
else:
area_look[info["position"]] += " " + info["description"]
info["description"] = area_look[info["position"]]
object_dict[info["position"]] = 1
#If the place being visited is a new place then the description is stored
else:
area_look[info["position"]] = info["description"]
#getting the training value for every 10 iterations
if(i%100 == 0):
#print("y=iterations : ",i)
batches = exp.return_seq()#exp.sample(num1=3,num2=5)
X_d,X_action,Y_qvalue,Y_statevalue = get_training(batches)
#print("X : {}, X_action : {}, Y_q : {}, Y_s : {}".format(X_d.shape,X_action.shape,Y_qvalue.shape,Y_statevalue.shape))
log_dir = "logs/" + datetime.now().strftime("%Y%m%d-%H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1)
model.fit(x = [X_d,X_action],y = [Y_qvalue,Y_statevalue],epochs=50,verbose=1,callbacks=[tensorboard_callback])
if i<1000 and epsilon_value>0.2:
epsilon_value-=0.2
# if reward is too less the episode terminates
if exp.get_cum_reward()<-50:
info = get_state_variables("restart",None)
i = 0
episode_count+=1
print("\nEpisode ",episode_count," Terminated\n")
epsilon_value = 0.8
i+=1
next_info,score = get_state_variables("restart",current_location.name)
epsilon_value = 0.8
print("\nEpisode ",episode_count," Terminated\n")
episode_count+=1
return None
if __name__ == '__main__':
graph = propogate()
while isinstance(graph,dict):
graph = propogate(model=graph["model"],exp=graph["exp"],score_dict=graph["score_dict"],episode=graph["episode"])
#Accessing elements of the graph
'''print("\nNodes : ",graph.n_nodes,"\n")
for i in graph.n_nodes:
print(i[1])'''