added second blackjack model

Blackjack.py

@@ -86,7 +86,7 @@ def hand( montecarlo, debug ):
 		# if hitting
 		if choice is "h":
 			# add data
-			data = data + [ hand_value( players_hand ) ]
+			data = data + [ [ hand_value( players_hand ), int( dealers_hand[0][:dealers_hand[0].index('-')] ) ] ]
 			# hit
 			players_hand = players_hand + [ deck.deal() ]
 			summ = hand_value( players_hand )
@@ -118,7 +118,7 @@ def hand( montecarlo, debug ):
 		# if staying
 		else:
 			# add data
-			data = data + [ hand_value( players_hand ) ]
+			data = data + [ [ hand_value( players_hand ), int( dealers_hand[0][:dealers_hand[0].index('-')] ) ] ]
 			if debug:
 				print( "Staying" )
 				# print current game
@@ -185,15 +185,15 @@ def hand( montecarlo, debug ):
 # data points to the blackjack data set
 def gen_data_set( ):
 	# loop through a thousand simulations
-	for i in range( 3000 ):
+	for i in range( 2000 ):
 		print( i )
 		data, tags = hand( True, False )
 
 		print( data )
 		print( tags )
 
-		dataf = open( "data_sets/blackjack.data.1", "a" )
-		tagf = open( "data_sets/blackjack.tags.1", "a" )
+		dataf = open( "data_sets/blackjack.data.2", "a" )
+		tagf = open( "data_sets/blackjack.tags.2", "a" )
 		for datum in data:
 			dataf.write( str( datum ) + "\n" )
 		for tag in tags:

README.md

@@ -1,5 +1,25 @@
 # card-games
- A collection of Python classes and scripts for using card games in Monte Carlo simulations and Machine Learning.
+A collection of Python classes and scripts for using card games in Monte Carlo simulations and Machine Learning.
+
+# Deck.py
+Deck.py is a class for a deck of cards. An instance of this class holds an array of 52 strings representing cards. Card strings consist of the face value and a suit seperated by a hyphen. Face values { A, 2, ... Q, K } are represented as { 1, 2, ... 12, 13 }. Suits are { 's', 'd', 'c', 'h' }. Some examples of valid card strings are '1-s' for the ace of spades, '4-c' for the four of clubs, and '12-h' for the queen of hearts.
+
+Usage:
+
+from Deck import Deck
+deck = Deck()
+
+Functions
+
+deck.shuffle()
+returns the array to 52 randomly organized card strings.
+
+deck.deal()
+removes a card string from the Deck object, and returns it
+
+deck.checkDeck()
+prints the current content of the deck
+
 
 # blackjack.py
 Blackjack.py is an implementation of the classic card game using the Deck class, and played from the terminal. This script is used to generate data about hands of blackjack via Monte Carlo simulations. This is done by generating random hands and storing representations of the hands, tagged with the eventual outcome of the decision.
@@ -109,21 +129,16 @@ The model learned to hit on any hand value below 17. This happens to be the stra
 
 Future data sets and models will focus learning more about the hand, ie. dealers cards, counting how many times a player has hit, etc. The longer term goal is to teach a model to count cards.
 
-# Deck.py
-Deck.py is a class for a deck of cards. An instance of this class holds an array of 52 strings representing cards. Card strings consist of the face value and a suit seperated by a hyphen. Face values { A, 2, ... Q, K } are represented as { 1, 2, ... 12, 13 }. Suits are { 's', 'd', 'c', 'h' }. Some examples of valid card strings are '1-s' for the ace of spades, '4-c' for the four of clubs, and '12-h' for the queen of hearts.
+# second Blackjack model
 
-Usage:
+This model will use all the previous techniques, but the data set will now include the dealer's upward facing card.
 
-from Deck import Deck
-deck = Deck()
+So where previously we used the single integer for data, we will now use a tuple of players_hand, and dealers_hand respectively.
 
-Functions
-
-deck.shuffle()
-returns the array to 52 randomly organized card strings.
+Example
 
-deck.deal()
-removes a card string from the Deck object, and returns it
+[ 18, 13, s ]
 
-deck.checkDeck()
-prints the current content of the deck
+The data set consists of 5,323 entries, located in data_sets/blackjack.data.2 and data_sets/blackjack.tags.2.. The neural network used a similar layer scheme as the previous, with an 8-neuron second layer. The optimizer was 'nadam,' and there were 100 epochs.
+
+The model has an accuracy of 0.74. A slightly better accuracy than model #1.