def heatmap(eta_vals, lmbd_vals, X_test_sc, X_train_sc, Y_train_onehot, y_test,
epochs, batch_size, n_hidden_neurons, n_categories):
accuracy_array = np.zeros((len(eta_vals), len(lmbd_vals)), dtype=object)
# grid search
for i, eta in enumerate(eta_vals):
for j, lmbd in enumerate(lmbd_vals):
dnn = NN(X_train_sc,
Y_train_onehot,
eta=eta,
lmbd=lmbd,
epochs=epochs,
batch_size=batch_size,
n_hidden_neurons=n_hidden_neurons,
n_categories=n_categories)
dnn.train()
test_predict = dnn.predict(X_test_sc)
accuracy_array[i][j] = accuracy_score(y_test, test_predict)
print("Learning rate = ", eta)
print("Lambda = ", lmbd)
print("Accuracy score on test set: ",
accuracy_score(y_test, test_predict))
print()
np.save('acc_score', accuracy_array)
np.save('eta_values', eta_vals)
np.save('lambda_values', lmbd_vals)
P.map()
def __init__(self, map, x, y, sim, model=None, train=False):
Bot.__init__(self, map, x, y, sim)
if model:
self.model = model
else:
self.model = NN(33, [10, 8])
# attribue le type au bot
self.type = "B" # => bot normal
self.train = train
if train: # si le bot est en mode training
# lui donne de l'energie en plus pour avoir un peux plus de temps pour train
self.incr_energy(100)
# lui donne le type training
self.type = "T"
self.cellNum = 0
self.e_fruit = 4 # energy donne en absorbant un fruit
self.e_meat = 20 # energy donne en absorbant de la viande
self.c_move = 1 # cout de bouger
self.c_rien = 1 # cout de rien faire
class Gene:
def __init__(self, input_len, hidden_layers, output_len):
self.input_len = input_len
self.hidden_layers = hidden_layers
self.output_len = output_len
self.nn = NN(input_len, hidden_layers, output_len)
def breed(self, gene): # weight-wise? # layer-wise? # average everything?
# print(gene)
# technically not the best way of doing this because it will first init the weights and thus wastes time...
new_gene = Gene(self.input_len, self.hidden_layers, self.output_len)
for i in np.arange(len(self.nn.weights)):
new_gene.nn.weights[i] = (self.nn.weights[i] +
gene.nn.weights[i]) / 2
for i in np.arange(len(self.nn.biases)):
new_gene.nn.biases[i] = (self.nn.biases[i] + gene.nn.biases[i]) / 2
return new_gene
def mutate(self): # add epsilon? # pick a weight and re-init?
pass
def action(self, state):
return self.nn.predict(state)
def save(self, file_name):
self.nn.save(file_name)
def load(self, file_name):
self.nn.load(file_name)
def train(config):
'''The config object required for training is in train_config.py'''
config['hyperparameters']['seed'] = np.random.randint(1e5)
if config['dataset_name'] == 'mnist':
dataset = load_mnist(config['mnist_sample_shape'])
elif config['dataset_name'] == 'cifar10':
dataset = load_cifar10(flatten_input=(config['nn_type'] == 'mlp'))
else:
raise ValueError(err_msg)
nn = NN(data=dataset, **config['hyperparameters'])
start = timeit.default_timer()
train_logs = nn.train_loop(eval_each_epoch=config['eval_while_training'])
elapsed = round(timeit.default_timer()-start,4)
print(f'training runtime: {elapsed} seconds')
exp = utils.ExperimentResults()
exp.save(train_logs, 'train_logs')
exp.save(config, 'config')
exp.save(nn, 'neural_network')
exp.save(elapsed,'training_runtime')
test_results = nn.evaluate()
exp.save(test_results, 'test_results')
def __init__(self, z_dim, g_dim, x_dim, h_dim, lr=0.01, dropout=0.0):
#结构参数
self.z_dim = z_dim
self.g_dim = g_dim
self.x_dim = x_dim
self.h_dim = h_dim
#学习参数
assert (lr > 0)
self.lr = float(lr)
self.dropout = min(max(dropout, 0.0), 1.0)
##要为生成器和判别器分别分配优化器
self.generator_optimizer = Adam(alpha=self.lr)
#self.discriminator_optimizer = Adam(alpha=self.lr)
#网络参数
self.generator = NN(input_dim=z_dim,
hidden_dim=g_dim,
output_dim=x_dim,
lr=self.lr * 0.5,
dropout=self.dropout)
self.generator.mode = 'binary'
self.discriminator = NN(input_dim=x_dim,
hidden_dim=h_dim,
output_dim=1,
lr=self.lr * 0.1,
dropout=self.dropout)
def mnist_net():
mndata = MNIST("mnist", return_type="numpy")
print("Loading images...")
images, labels = mndata.load_training()
features = images.T / 255
z = np.zeros((60000, 10))
z[np.arange(60000), labels] = 1
Y = z.T
nn = NN([784, 100, 30, 10])
nn.set_hyperparameters(learning_rate=0.5)
t = time()
nn.initialize_parameters()
print("Start Training...")
nn.minimize({"features": features, "labels": Y}, 20)
print("Finish Training.")
print("Training time: {0} seconds".format(round(time() - t, 2)))
print("Start Testing...")
t = time()
test_images, test_labels = mndata.load_testing()
test_features = test_images.T / 255
z = np.zeros((10000, 10))
z[np.arange(10000), test_labels] = 1
test_Y = z.T
print("Testing accuracy: {}".format(
round(nn.evaluate({
"features": test_features,
"labels": test_Y
}), 4)))
print("Testing time: {0} seconds".format(round(time() - t, 2)))
def __init__(self):
info_df = pd.read_csv('hdb-carpark-information-with-lat-lng.csv')
self.nn = NN()
df = self.nn.getCurrentAvailability()
self.info_df = pd.merge(df,
info_df,
left_on=['carpark_number'],
right_on=['carpark_number'])
self.parkings = self.info_df[[
'carpark_number', 'lat', 'lng', 'night_parking', 'free_parking',
'car_park_type', 'type_of_parking_system'
]]
self.db = firestore.client()
def __init__(self, x_dim, h_dim, z_dim, g_dim, lr=0.01, dropout=0.0):
#结构参数
self.z_dim = z_dim
self.h_dim = h_dim
self.g_dim = g_dim
self.x_dim = x_dim
#学习参数
assert (lr > 0)
self.lr = float(lr)
self.dropout = min(max(dropout, 0.0), 1.0)
#网络参数
self.encoder = SimpleNN(input_dim=x_dim,
hidden_dim=h_dim,
output_dim=2 * z_dim,
lr=self.lr,
dropout=self.dropout)
self.encoder_optimizer = Adam(alpha=self.lr)
self.decoder = NN(input_dim=z_dim,
hidden_dim=g_dim,
output_dim=x_dim,
lr=self.lr,
dropout=self.dropout)
self.decoder.mode = 'binary'
def __init__(self, map, x, y, sim, model=None):
self.map = map
self.x = x
self.y = y
self.incr_energy(10)
self.nb_steps = 0
self.sim = sim
self.model = NN(13, [20, 8])
self.type = "A"
def getFilteredParkings(self, lat, lon, night_parking, free_parking,
car_park_type, type_of_parking_system):
nn = NN()
df = nn.getCurrentAvailability()
df = pd.merge(df,
self.parkings,
left_on=['carpark_number'],
right_on=['carpark_number'])
id = self.DistCalc(lat, lon)
# filtparkings=df[(df['night_parking']==night_parking) & (df['free_parking']==free_parking) &(df['car_park_type'].isin(car_park_type)) & (df['type_of_parking_system']==type_of_parking_system)]
# filteredparkings=filtparkings[filtparkings['carpark_number'].isin(id)]
# print(filteredparkings)
# res=filteredparkings[['carpark_number','lat','lng','lots_available']].set_index('carpark_number').T.to_json()
filtparkings = df
if night_parking != None:
filtparkings = filtparkings[(
filtparkings['night_parking'] == night_parking)]
if free_parking != None:
filtparkings = filtparkings[(
filtparkings['free_parking'] == free_parking)]
if car_park_type != None:
filtparkings = filtparkings[(
filtparkings['car_park_type'].isin(car_park_type))]
if type_of_parking_system != None:
filtparkings = filtparkings[(filtparkings['type_of_parking_system']
== type_of_parking_system)]
filteredparkings = filtparkings[filtparkings['carpark_number'].isin(
id)]
#filtparkings=df[(df['night_parking']==req.get("night_parking")) & (df['free_parking']==req.get("free_parking")) & (df['car_park_type'].isin(req.get("car_park_type"))) & (df['type_of_parking_system']==req.get("type_of_parking_system"))]
print(filteredparkings)
res = filteredparkings[[
'carpark_number', 'lat', 'lng', 'lots_available'
]].set_index('carpark_number').T.to_json()
return res
def train(hyperparameters, sample_shape, nn_type):
hyperparameters['seed'] = np.random.randint(1e5)
cifar10 = load_cifar10(flatten_input=(
nn_type == 'mlp')) # because mlp only processes 1D input
nn = NN(data=cifar10, **hyperparameters)
perform_evaluation = False
if nn_type == 'mlp':
perform_evaluation = True
elif hyperparameters['n_epochs'] == 1:
perform_evaluation = True
train_logs = nn.train_loop(eval_each_epoch=perform_evaluation)
exp = utils.ExperimentResults()
exp.save(train_logs, 'train_logs')
exp.save(hyperparameters, 'hyperparams')
exp.save(nn, 'neural_network')
test_results = nn.evaluate()
exp.save(test_results, 'test_results')
def xor_net():
a = np.array([[0, 0, 0], [0, 1, 1], [1, 0, 1], [1, 1, 0]])
features = a[:, 0:2].T
labels = a[:, 2].reshape((1, 4))
z = np.zeros((4, 2))
z[np.arange(4), labels] = 1
labels = z.T
nn = NN([2, 4, 3, 2])
nn.set_hyperparameters(batch_size=4, learning_rate=0.75)
t = time()
nn.initialize_parameters()
print("Start Training...")
nn.minimize({"features": features, "labels": labels}, 10000)
print("Finish Training.")
print("Training time: {0} seconds".format(round(time() - t, 2)))
print("Start Testing...")
t = time()
print("Testing accuracy: {}".format(
round(nn.evaluate({
"features": features,
"labels": labels
}), 4)))
print("Testing time: {0} seconds".format(round(time() - t, 2)))
def gen_NN(genes=[]):
# # Inputs
# input = Input(shape=(7,))
# x = Dense(10, activation='tanh')(input)
# x = Dense(10, activation='tanh')(x)
# predictions = Dense(2, activation='tanh')(x)
# model = Model(inputs=input, outputs=predictions)
# if len(genes) > 0:
# model.set_weights(genes)
# model._make_predict_function()
model = NN()
return model
def __init__(self,
z_dim,
g_dim,
x_dim,
h_dim,
n_classes,
embedding_dim=8,
lr=0.01,
dropout=0.0):
self.n_classes = n_classes
new_z_dim = z_dim + embedding_dim
super(AcGAN, self).__init__(new_z_dim, g_dim, x_dim, h_dim, lr,
dropout)
self.classifier = NN(input_dim=x_dim,
hidden_dim=h_dim,
output_dim=n_classes,
lr=self.lr,
dropout=self.dropout)
#将条件变量嵌入到一个向量空间，并保持为常量
self.c_vectors = normal_sample(dim=embedding_dim, n=n_classes)
self.embedding_dim = embedding_dim
def __init__(self):
# appropriately initialize your neural network
self.model = NN([2, 1])
#!/bin/python
import numpy as np
from neural_network import NN
xor_states = np.array( [ [0,0], [0,1], [1,1], [1,0] ] )
xor_outputs = np.array( [ 0, 1, 0, 1 ] )
net = NN()
net.load()
net.useThresholds = 1
net.useSigmoid = 1
for inpt, out in zip( xor_states, xor_outputs ):
print net.run( inpt )
print out
print "\n\n"
#from __future__ import print_function as print
from neural_network import NeuralNetwork as NN
import torch
from logic_gates import AND, OR, NOT, XOR
if __name__ == "__main__":
nn = NN([2, 1])
and_gate = AND()
or_gate = OR()
not_gate = NOT()
xor_gate = XOR()
print(and_gate(False, False))
print(and_gate(False, True))
print(and_gate(True, False))
print(and_gate(True, True))
print(" ")
print(or_gate(False, False))
print(or_gate(False, True))
print(or_gate(True, False))
print(or_gate(True, True))
print("")
print(not_gate(True))
print(not_gate(False))
print("")
print(xor_gate(False, False))
print(xor_gate(False, True))
def __init__(self, input_len, hidden_layers, output_len):
self.input_len = input_len
self.hidden_layers = hidden_layers
self.output_len = output_len
self.nn = NN(input_len, hidden_layers, output_len)
# Iris Mutli-Class Classifier
"""
NN class takes in a model, input training data and target data
as arguments.
The model argument is a Python list that defines the layers in a
neural network. The length of the defines how 'deep' the neural_network is,
Each element of the list defines the number of neurons at the given layer.
ex: model = [4, 5, 7, 3] |
This model is a neural network with 4 layers
it has 4 input neurons or features and 3 output neurons(classes)
The 1st hidden layer has 5 neurons; the 2nd hidden layer has 7 neurons
"""
NN1 = NN(model_1, X_train_1, Y_train_1)
NN1.activation_function('tanh')
NN1.output_function('softmax')
NN1.loss_function('cross_entropy')
# Breast Cancer Logistic Regression
NN2 = NN(model_2, X_train_2, Y_train_2)
NN2.activation_function('tanh')
NN2.output_function('sigmoid')
NN2.loss_function('log_loss')
# Train neural networks
NN1.train()
NN1.predict()
NN2.train()
NN2.predict()
class Parking:
def __init__(self):
info_df = pd.read_csv('hdb-carpark-information-with-lat-lng.csv')
self.nn = NN()
df = self.nn.getCurrentAvailability()
self.info_df = pd.merge(df,
info_df,
left_on=['carpark_number'],
right_on=['carpark_number'])
self.parkings = self.info_df[[
'carpark_number', 'lat', 'lng', 'night_parking', 'free_parking',
'car_park_type', 'type_of_parking_system'
]]
self.db = firestore.client()
def initializeLocations(self):
#for every record in the df, store it in the parking_info collection using the parking ID as document ID
parking = self.info_df.set_index('carpark_number').T.to_json()
parking = json.loads(parking)
#sending the array to firestored
data = {u'parking': parking}
self.db.collection(u'parkingsinfo').document('parkings').set(data)
#return true after you're done
return (True, )
def updateCurrentAvailability(self):
next_call = time.time()
while True:
print("getting current availability")
parking = self.nn.getCurrentAvailability()
parking = self.info_df.set_index('carpark_number')[[
'lots_available'
]].T.to_json()
parking = json.loads(parking)
data = {u'current_availability': parking}
self.db.collection(u'parking_info').document('parkings').set(data)
print("pushed to db")
next_call += 60
print("sleeping")
time.sleep(next_call - time.time())
def DistCalc(self, latitude, longtitude):
R = 6373.0
location = []
lattemp = latitude
lontemp = longtitude
j = 0
for i in range(len(self.parkings.lat)):
lat1 = radians(self.parkings.lat[i])
lon1 = radians(self.parkings.lng[i])
lat2 = radians(lattemp)
lon2 = radians(lontemp)
dlon = lon2 - lon1
dlat = lat2 - lat1
a = sin(dlat / 2)**2 + cos(lat1) * cos(lat2) * sin(dlon / 2)**2
c = 2 * atan2(sqrt(a), sqrt(1 - a))
distance = R * c
if (distance < 1.0000):
location.append(self.parkings.carpark_number[i])
print(location[j])
j += 1
return location
def getFilteredParkings(self, lat, lon, night_parking, free_parking,
car_park_type, type_of_parking_system):
nn = NN()
df = nn.getCurrentAvailability()
df = pd.merge(df,
self.parkings,
left_on=['carpark_number'],
right_on=['carpark_number'])
id = self.DistCalc(lat, lon)
# filtparkings=df[(df['night_parking']==night_parking) & (df['free_parking']==free_parking) &(df['car_park_type'].isin(car_park_type)) & (df['type_of_parking_system']==type_of_parking_system)]
# filteredparkings=filtparkings[filtparkings['carpark_number'].isin(id)]
# print(filteredparkings)
# res=filteredparkings[['carpark_number','lat','lng','lots_available']].set_index('carpark_number').T.to_json()
filtparkings = df
if night_parking != None:
filtparkings = filtparkings[(
filtparkings['night_parking'] == night_parking)]
if free_parking != None:
filtparkings = filtparkings[(
filtparkings['free_parking'] == free_parking)]
if car_park_type != None:
filtparkings = filtparkings[(
filtparkings['car_park_type'].isin(car_park_type))]
if type_of_parking_system != None:
filtparkings = filtparkings[(filtparkings['type_of_parking_system']
== type_of_parking_system)]
filteredparkings = filtparkings[filtparkings['carpark_number'].isin(
id)]
#filtparkings=df[(df['night_parking']==req.get("night_parking")) & (df['free_parking']==req.get("free_parking")) & (df['car_park_type'].isin(req.get("car_park_type"))) & (df['type_of_parking_system']==req.get("type_of_parking_system"))]
print(filteredparkings)
res = filteredparkings[[
'carpark_number', 'lat', 'lng', 'lots_available'
]].set_index('carpark_number').T.to_json()
return res
def generateSequencePrediction(self):
next_call = time.time()
while True:
df_list = self.nn.getSequenceFromCurrentTime()
self.df_list_for_nn = self.nn.modifyDataframeListForNN(df_list)
predictions = {}
#generate prediction
for df in self.df_list_for_nn:
model = self.models[df.carpark_number.iloc[0]]
prediction = self.nn.generatePrediction(model, df)
prediction = prediction[0].tolist()
predictions[df.carpark_number.iloc[0]] = prediction
print("[", df.carpark_number.iloc[0], "] = ", prediction)
#push to firebase
data = {u'predictions': predictions}
self.db.collection(u'parking_predictions').document(
'predictions').set(data)
next_call += 60 * 15 # every 15 minutes
time.sleep(next_call - time.time())
def loadModels(self):
#read all the models from the folder
#use this:
self.models = {}
directory = r"./trained_models/"
for filename in os.listdir(directory):
#load model
model = load_model(directory + filename,
custom_objects=None,
compile=True,
options=None)
#get carpark id from filename
filename_split = filename.split('_')
carpark_id_split = filename_split[-1].split('.')
carpark_id = carpark_id_split[0]
#add model to dict with car_park id as key
self.models[carpark_id] = model
print("[" + carpark_id + "]=\t")
print(self.models[carpark_id])
# plotting single number from training
def show_example(data, x):
img_array = np.asfarray(data[x].split(',')[1:]).reshape((28,28))
plt.imshow(img_array, cmap='Greys', interpolation='None')
plt.show()
# Number of input, hidden, and output nodes
input_nodes = 784
hidden_nodes = 100
output_nodes = 10
# Learning rate
learning_rate = 0.1
# Initialise Neural network
N = NN(input_nodes, hidden_nodes, output_nodes, learning_rate)
# TRAINING
epochs = 2
for epoch in range(epochs):
for sample in training_data:
sample_values = sample.split(',')
# scale and shift input
inputs = (np.asfarray(sample_values[1:]) / 255.0 * 0.99) + 0.01
# create target values (0.01 -> 0.99)
targets = np.zeros(output_nodes) + 0.01
# assign desired target values
targets[int(sample_values[0])] = 0.99
N.train(inputs, targets)
import tensorflow as tf
from neural_network import NN
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
y_train = tf.one_hot(y_train, depth=10).numpy()
y_test = tf.one_hot(y_test, depth=10).numpy()
x_train = x_train.reshape(x_train.shape[0], -1).T
y_train = y_train.reshape(y_train.shape[0], -1).T
dev_X = x_test.reshape(x_test.shape[0], -1).T
dev_Y = y_test.reshape(y_test.shape[0], -1).T
nn = NN([784, 10, 40, 20, 10])
x = tf.Variable(x_train, dtype='float32')
y = tf.Variable(y_train, dtype='float32')
nn.model(x, y, dev_X, dev_Y, num_epochs=100, learning_rate=0.001, minibatch_size=512, beta_1=0.9, lamda=0,
softmax=True)
n_categories = 2
eta_vals = np.logspace(-7, -4, 7)
lmbd_vals = np.logspace(-7, -1, 7)
# Make heatmap of the accuracy score with eta_vals and lmbd_vals
# Commented out to save time
#func.heatmap(eta_vals, lmbd_vals, X_test_sc, X_train_sc, Y_train_onehot, y_test, epochs, batch_size, n_hidden_neurons, n_categories)
# Use best values from heatmap
eta_final = 1e-4
lmbd_final = 1e-2
dnn_f = NN(X_train_sc,
Y_train_onehot,
eta=eta_final,
lmbd=lmbd_final,
epochs=epochs,
batch_size=batch_size,
n_hidden_neurons=n_hidden_neurons,
n_categories=n_categories)
dnn_f.train()
y_predict = dnn_f.predict(X_test_sc)
model = y_predict
# Make Cumulative gain plot
P.Cumulative_gain_plot(y_test, model)
# Creating a Confusion matrix using pandas and pandas dataframe
CM = func.Create_ConfusionMatrix(model, y_test, plot=True)
CM_DataFrame = func.ConfusionMatrix_DataFrame(
CM, labels=['pay', 'default'])
class NN_bot(Bot):
def __init__(self, map, x, y, sim, model=None, train=False):
Bot.__init__(self, map, x, y, sim)
if model:
self.model = model
else:
self.model = NN(33, [10, 8])
# attribue le type au bot
self.type = "B" # => bot normal
self.train = train
if train: # si le bot est en mode training
# lui donne de l'energie en plus pour avoir un peux plus de temps pour train
self.incr_energy(100)
# lui donne le type training
self.type = "T"
self.cellNum = 0
self.e_fruit = 4 # energy donne en absorbant un fruit
self.e_meat = 20 # energy donne en absorbant de la viande
self.c_move = 1 # cout de bouger
self.c_rien = 1 # cout de rien faire
#TODO rajouter et gerer les variables pour la mitose
def g_inputs(self):
# TODO mettre en log les energis
inputs = np.array([])
inputs = np.append(inputs, np.cbrt(self.g_energy()))
inputs = np.append(inputs, np.cbrt(self.g_nb_fruit_on_pos()))
# des inputs qui servent juste a donner des variables qui bouclent pour
# permettre au bot un peu de changement dans son comportement et lui
# permettre d'avoir des actions un peu cyclique
inputs = np.append(inputs, np.cbrt(self.sim.current_nb_step % 2))
inputs = np.append(inputs, np.cbrt(self.sim.current_nb_step % 10))
inputs = np.append(inputs, np.cbrt(self.sim.current_nb_step % 50))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 6, [0, 1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 6, [0, -1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 6, [1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 6, [-1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 5, [0, 1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 5, [0, -1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 5, [1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 5, [-1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 4, [0, 1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 4, [0, -1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 4, [1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 4, [-1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 3, [0, 1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 3, [0, -1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 3, [1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 3, [-1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 2, [0, 1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 2, [0, -1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 2, [1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 2, [-1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 1, [0, 1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 1, [0, -1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 1, [1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 1, [-1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 0, [0, 1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 0, [0, -1])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 0, [1, 0])))
inputs = np.append(inputs, np.cbrt(self.g_info_sum_on_dir(21, 0, [-1, 0])))
# inputs = np.append(inputs, self.g_nb_fruit_on_dir([0, 1], 3))
# inputs = np.append(inputs, self.g_nb_fruit_on_dir([0, -1], 3))
# inputs = np.append(inputs, self.g_nb_fruit_on_dir([1, 0], 3))
# inputs = np.append(inputs, self.g_nb_fruit_on_dir([-1, 0], 3))
# inputs = np.append(inputs, self.g_bot_on_dir([1, 0]))
# inputs = np.append(inputs, self.g_bot_on_dir([-1, 0]))
# inputs = np.append(inputs, self.g_bot_on_dir([0, 1]))
# inputs = np.append(inputs, self.g_bot_on_dir([0, -1]))
# inputs = np.append(inputs, self.g_bot_on_dir([1, 0], 3))
# inputs = np.append(inputs, self.g_bot_on_dir([-1, 0], 3))
# inputs = np.append(inputs, self.g_bot_on_dir([0, 1]), 3)
# inputs = np.append(inputs, self.g_bot_on_dir([0, -1], 3))
return inputs
def predict(self):
inputs = self.g_inputs()
prediction = self.model.predict(inputs)
action = np.argmax(prediction)
if self.train:
albot_actions = self.albot_predict()
self.model.fit_on_one(self.g_inputs(), albot_actions, 0.001)
action = albot_actions
return action
def albot_predict(self):
actions = [0, 0, 0, 0, 0, 0, 0, 0]
if self.g_energy() > 300:
pass
# actions[6] = 1000
if self.g_nb_fruit_on_pos() > 0:
actions[7] = 10000
actions[1] = self.g_nb_fruit_on_dir([0, -1], 3) - (
self.g_bot_on_dir([0, -1]) * 0000
)
actions[2] = self.g_nb_fruit_on_dir([0, 1], 3) - (
self.g_bot_on_dir([0, 1]) * 0000
)
actions[3] = self.g_nb_fruit_on_dir([-1, 0], 3) - (
self.g_bot_on_dir([-1, 0]) * 0000
)
actions[4] = self.g_nb_fruit_on_dir([1, 0], 3) - (
self.g_bot_on_dir([1, 0]) * 0000
)
return actions
def mitose(self):
if self.g_cd_repro() == 0:
self.incr_cd_repro(20)
self.incr_energy(-5) # loose of energy to make the child
# energy that will be transfered to the child
energy_to_child = 5
x = -1
y = -1
# TODO: faire une fonction pour rendre ca plus propre
if self.y - 2 > 0:
if self.map.cellLibre(self.x, self.y-1) == 0:
x = self.x
y = self.y - 1
elif self.y + 2 < self.map.height - 1:
if self.map.cellLibre(self.x, self.y+1) == 0:
x = self.x
y = self.y + 1
elif self.x - 2 > 0:
if self.map.cellLibre(self.x-1, self.y) == 0:
x = self.x - 1
y = self.y
elif self.x + 2 < self.map.height - 1:
if self.map.cellLibre(self.x+1, self.y) == 0:
x = self.x + 1
y = self.y
else:
# no place to put the child
self.incr_energy(energy_to_child)
self.incr_energy(-1)
if x == -1:
pass
else:
new_model = genetic.mutate(self.model.weights, 1, 1)
new_bot = NN_bot(self.map, x, y, self.sim, new_model)
new_bot.s_energy(energy_to_child)
self.sim.add_bots([new_bot])
