def adi(iterations=10000):
model = None
if os.path.exists(model_path):
model = keras.models.load_model(model_path)
else:
model = buildModel(20*24)
compile_model(model, 0.001)
for it in range(iterations):
# generate N scrambled cubes
l = 100
k = 20
cube_agent = CubeAgent(number_of_cubes=l, batches=k)
cube_agent.scramble_cubes_for_data()
cubes = np.array(cube_agent.env).flatten()
#initialize the training parameters -> marked by X and Y in the paper
encodedStates = np.empty((k*l, 20*24))
values = np.empty((k*l, 1))
policies = np.empty(k*l)
# iterate through the number of cubes and the number of actions
i = 0
for stateNumber , state in enumerate(cubes):
valuesForState = np.zeros(len(state.action_space))
encodedStates[i] = np.array(state.get_one_hot_state().flatten())
actions = state.action_space
start_state = state.cube.copy()
#1-depth BFS
for j, action in enumerate(actions):
_ , reward = state.step(j)
childStateEncoded = np.array(state.get_one_hot_state()).flatten()
state.set_state(start_state) #set back to the original
value, policy = model.predict(childStateEncoded[None, :])
valueNumber = value[0][0]
valuesForState[j] = valueNumber + reward
values[i] = np.array([valuesForState.max()])
policies[i] = valuesForState.argmax()
i += 1
#log into Tensorboad
log_dir = "logs/fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1)
sample_weight = np.array([[1/(i+1) for i in range(k)] for j in range(l)]).flatten()
model.fit(encodedStates,
{ "output_policy": policies, "output_value": values },
epochs=15, sample_weight=sample_weight)
model.save(model_path)
return model
def main(args):
# Loads the dataset
X_TRAIN = pickle.load(
open(f"{args['data_destination']}/X_TRAIN.pickle", "rb"))
Y_TRAIN = pickle.load(
open(f"{args['data_destination']}/Y_TRAIN.pickle", "rb"))
#X_TEST = pickle.load(open(f"{args['data_destination']}/X_TEST.pickle", "rb"))
#Y_TEST = pickle.load(open(f"{args['data_destination']}/Y_TEST.pickle", "rb"))
#X_VALIDATION = pickle.load(open(f"{args['data_destination']}/X_VALIDATION.pickle", "rb"))
#Y_VALIDATION = pickle.load(open(f"{args['data_destination']}/Y_VALIDATION.pickle", "rb"))
#normalize
X_TRAIN = X_TRAIN / 255.0
#X_TEST = X_TEST/255.0
#X_VALIDATION = X_VALIDATION/255.0
dct_model = model.buildModel(
tuple(int(dim) for dim in args['shape'].split(',')),
int(args['classes']))
model.train(dct_model, X_TRAIN, Y_TRAIN, int(args['batch_size']),
int(args['epochs']))
model_sum = dct_model.summary()
print(f'model summary: {model_sum}')
# model.eval(dct_model, X_TEST, Y_TEST, int(args['batch_size']))
try:
model.predict(dct_model, X_TEST)
except:
print("model.predict gave an error")
model.saveModel(dct_model)
def main():
data_directory = '/home/linux-box0520/data/food-101/data2/'
TRAIN_BATCHSIZE = 16
EVAL_BATCHSIZE = 1
data_transforms = {
'train':
transforms.Compose([
transforms.RandomResizedCrop(224),
#transforms.RandomApply([
# transforms.RandomHorizontalFlip(),
# transforms.RandomVerticalFlip(),
# transforms.ColorJitter(),
# transforms.RandomRotation((1,360))]),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
]),
'val':
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
]),
}
image_datasets = {
x: datasets.ImageFolder(os.path.join(data_directory, x),
data_transforms[x])
for x in ['train', 'val']
}
dataloaders = {
x: torch.utils.data.DataLoader(
image_datasets[x],
batch_size=(TRAIN_BATCHSIZE if x == 'train' else EVAL_BATCHSIZE),
shuffle=True,
num_workers=1)
for x in ['train', 'val']
}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']}
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = buildModel()
model = model.to(device)
loss = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=.005, momentum=.93)
model = train_model(model,
loss,
optimizer,
dataloaders,
dataset_sizes,
device,
num_epochs=200)
torch.save(model.state_dict(), 'deconvNet-200epochs-3classes.pt')
def playGame(observe=False):
tensorModel = buildModel()
game = Game()
agent = Agent(game)
train = Train(agent, game)
try:
train.trainNetwork(tensorModel, observe=observe)
except StopIteration:
game.End()
def runTournamentBracket(model=None):
year = conf.year
if model is None:
model = Model.buildModel(200)
# print ("Running Tournament")
west = Bracket.runbracket(conf.teams['west'], model)
east = Bracket.runbracket(conf.teams['east'], model)
south = Bracket.runbracket(conf.teams['south'], model)
midwest = Bracket.runbracket(conf.teams['midwest'], model)
final4teams = [west[4][0], east[4][0], south[4][0], midwest[4][0]]
final4 = Bracket.runbracket(final4teams, model)
champion = final4[2][0]
return {
'year': 2016,
'west': west,
'east': east,
'south': south,
'midwest': midwest,
'final4': final4,
'champion': champion
}
def loadModel(stateDictPath):
model = buildModel()
model.load_state_dict(torch.load(stateDictPath))
model.eval()
return model
def train_(base_path):
data, anno = read_data(base_path)
anno = np.expand_dims(anno, axis=-1)
mean = np.mean(data)
std = np.std(data)
data_ = (data - mean) / std
train_data = data_[:150]
train_anno = anno[:150]
val_data = data_[150:]
val_anno = anno[150:]
print('-' * 30)
print(
'Creating and compiling the fully convolutional regression networks.')
print('-' * 30)
model = buildModel(input_dim=(256, 256, 3))
model_checkpoint = ModelCheckpoint('cell_counting.hdf5',
monitor='loss',
save_best_only=True)
model.summary()
print('...Fitting model...')
print('-' * 30)
change_lr = LearningRateScheduler(step_decay)
datagen = ImageDataGenerator(
featurewise_center=False, # set input mean to 0 over the dataset
samplewise_center=False, # set each sample mean to 0
featurewise_std_normalization=
False, # divide inputs by std of the dataset
samplewise_std_normalization=False, # divide each input by its std
zca_whitening=False, # apply ZCA whitening
rotation_range=
30, # randomly rotate images in the range (degrees, 0 to 180)
width_shift_range=
0.3, # randomly shift images horizontally (fraction of total width)
height_shift_range=
0.3, # randomly shift images vertically (fraction of total height)
zoom_range=0.3,
shear_range=0.,
horizontal_flip=True, # randomly flip images
vertical_flip=True, # randomly flip images
fill_mode='constant',
dim_ordering='tf')
# Fit the model on the batches generated by datagen.flow().
model.fit_generator(
datagen.flow(train_data, train_anno, batch_size=16),
samples_per_epoch=train_data.shape[0],
nb_epoch=96,
callbacks=[model_checkpoint, change_lr],
)
model.load_weights('cell_counting.hdf5')
A = model.predict(val_data)
mean_diff = np.average(
np.abs(np.sum(np.sum(A, 1), 1) -
np.sum(np.sum(val_anno, 1), 1))) / (100.0)
print('After training, the difference is : {} cells per image.'.format(
np.abs(mean_diff)))
np.save(outputDir + "e_files_trainBatches", e_data[0])
np.save(outputDir + "e_events_trainBatches", e_data[1])
np.save(outputDir + "e_files_valBatches", e_data[2])
np.save(outputDir + "e_events_valBatches", e_data[3])
np.save(outputDir + "bkg_files_trainBatches", bkg_data[0])
np.save(outputDir + "bkg_events_trainBatches", bkg_data[1])
np.save(outputDir + "bkg_files_valBatches", bkg_data[2])
np.save(outputDir + "bkg_events_valBatches", bkg_data[3])
# initialize generators
train_generator = generator(e_data[0], e_data[1], bkg_data[0], bkg_data[1],
batch_size, dataDir, False, True, False)
val_generator = generator(e_data[2], e_data[3], bkg_data[2], bkg_data[3],
batch_size, dataDir, False, True, False)
model = buildModel()
model.compile(optimizer=optimizers.Adam(),
loss='categorical_crossentropy',
metrics=metrics)
callbacks = [
callbacks.EarlyStopping(monitor=monitor, patience=patience_count),
callbacks.ModelCheckpoint(filepath=weightsDir + 'model.{epoch}.h5',
save_best_only=True,
monitor=monitor,
mode='auto')
# tf.keras.callbacks.TensorBoard(log_dir=logDir,
# histogram_freq=0,
# write_graph=False,
# write_images=False)
if movie == '' or movie == None:
print('Error. Usage: python index.py --movie "<movie>"')
sys.exit()
try:
numberOfTweets = 1
tweets = []
ratingValue = 0
if model.isModelExists():
print("Loading Classifier Model...")
vectorizer, classifier = model.loadModel()
else:
print("Building Classifier Model...")
# Start building models
vectorizer, classifier = model.buildModel()
print("Retrieving Tweets...")
tso = TwitterSearchOrder()
tso.set_keywords([movie])
tso.set_language('en')
tso.set_include_entities(False)
ts = TwitterSearch(
consumer_key='kbibzVdoRoKOwd3dlxZCobum5',
consumer_secret='qEz32mJANlQ5hbFGacxmMfO2Pmyexs3WgPFeGq4QzA88qAOKe8',
access_token='1348353366-ofrMAMNiFfz102VY9c3MXdTrsAD2c4Dq91QiWVD',
access_token_secret='Io7orEnOvE2Rv2sdiASLRkLTVFA93DmSyF4r1i9CYQCXn')
for tweet in ts.search_tweets_iterable(tso):
preprocessed = preprocess.preprocess(tweet['text'])
tweets.append(preprocessed)
dest="load",
help="Load model matchup probabilities from file "
"(precalculated and saved using --save)",
action="store_true",
default=False)
parser.add_option("--save",
dest="save",
help="Save the calculated matchup probabilities to a file",
action="store_true",
default=False)
options, args = parser.parse_args()
model = None
if options.load:
model = Model.buildModel(load=True)
else:
model = Model.buildModel(nForests=options.nModels,
nTrees=options.nTrees,
save=options.save)
if options.calcStats:
TournamentStats.calculatePredictionStats(model, 100000)
elif options.makeBracket:
if options.winner:
TournamentStats.makeRandomBracket(model, options.winner)
else:
TournamentStats.makeRandomBracket(model)
elif options.predict:
team1 = options.predict[0]
team2 = options.predict[1]
"projectId": "sub-it-a29ca",
"storageBucket": "sub-it-a29ca.appspot.com",
"messagingSenderId": "744036348319",
"appId": "1:744036348319:web:8677ea4aeddccdda3fcdaa",
"measurementId": "G-STLRB5PMY1"
}
# Firebase integration
firebase = pyrebase.initialize_app(firebaseConfig)
auth = firebase.auth()
user = None
userID = None
db = firebase.database()
# Build ML model
modelVar = model.buildModel()
# Open ingredients JSON
with open("./alt_ingr.json", 'r') as fp:
alt_ingr = json.load(fp)
with open("./ingr_co2.json", 'r') as fp:
ingr_co2 = json.load(fp)
app = Flask(__name__)
@app.route("/")
def home():
# Redirect root to dashboard
return redirect(url_for("dash"))
def main():
trainDataFile = pd.read_csv(DATA_DIR + 'train_ship_segmentations_v2.csv')
print("Training Data (csv) File Shape:\t", trainDataFile.shape)
# Labeling NaN values
trainDataFile['NaN'] = trainDataFile['EncodedPixels'].apply(isNaN)
trainDataFile = trainDataFile.iloc[100000:]
trainDataFile = trainDataFile.sort_values('NaN', ascending=False)
print("\nNaN Value Count")
print(trainDataFile['NaN'].value_counts())
# Calculating Areas
trainDataFile['area'] = trainDataFile['EncodedPixels'].apply(calculateArea)
IsShip = trainDataFile[trainDataFile['area'] > 0]
train_group = trainDataFile.groupby('ImageId').sum()
print("\nGrouping Entries with same ImageId\nNaN Value Count")
print(trainDataFile['NaN'].value_counts())
train_group = train_group.reset_index()
# Assigning Classes
train_group['class'] = train_group['area'].apply(assignClasses)
print("\nClasses of Ships")
print(train_group['class'].value_counts())
# Train, Validation Split
trainingSet, validationSet = train_test_split(
train_group, test_size=0.01, stratify=train_group['class'].tolist())
print("\nTraining Set Shape:\t", trainingSet.shape[0])
print("Validation Set Shape:\t", validationSet.shape[0])
trainingSet_ship = trainingSet['ImageId'][trainingSet['NaN'] == 0].tolist()
trainingSet_nan = trainingSet['ImageId'][trainingSet['NaN'] == 1].tolist()
# Randomizing
trainingSet_ship = random.sample(trainingSet_ship, len(trainingSet_ship))
trainingSet_nan = random.sample(trainingSet_nan, len(trainingSet_nan))
EQUALIZED_DATA = min(len(trainingSet_ship), len(trainingSet_nan))
validationSet_ship = validationSet['ImageId'][validationSet['NaN'] ==
0].tolist()
validationSet_nan = validationSet['ImageId'][validationSet['NaN'] ==
1].tolist()
print("Training Set (Ships, Not Ships):\t", len(trainingSet_ship),
len(trainingSet_nan))
datagen = customGenerator(trainDataFile,
trainingSet_ship,
trainingSet_nan,
batchSize=BATCH_SIZE,
equalizedData=EQUALIZED_DATA)
valgen = customGenerator(trainDataFile,
validationSet_ship,
validationSet_nan,
batchSize=VALIDATION_BATCH_SIZE,
equalizedData=EQUALIZED_DATA)
validation_x, validation_y = next(valgen)
model = buildModel()
model.summary()
model.compile(optimizer=Adam(1e-3, decay=0.0),
metrics=['accuracy', f1],
loss='mean_squared_error')
model.save('ship_imageProcessing.h5')
# model = load_model('ship.h5', custom_objects={"f1" : f1})
# Training
history = model.fit_generator(datagen,
steps_per_epoch=250,
epochs=NUM_EPOCHS,
verbose=1,
validation_data=(validation_x, validation_y))
plt.subplot(2, 1, 1)
plt.plot(history.history['acc'])
plt.plot(history.history['val_acc'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Test'], loc='lower right')
plt.subplot(2, 1, 2)
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('Model Loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Train', 'Test'], loc='upper right')
plt.tight_layout()
plt.savefig('../Images/ModelPerformance.png')
plt.show()
predictions = model.predict(validation_x)
# print("First Prediction: ", predictions[0])
score = model.evaluate(validation_x, validation_y, verbose=1)
print('Validation loss:', score[0])
print('Validation accuracy:', score[1])
def main():
parser = argparse.ArgumentParser(
description=
"Train a bidirectional LSTM model for text sentiment classification")
parser.add_argument("--train_mode",
type=str,
default='preset_param',
choices=['preset_param', 'kerastuner'],
help="Set the training mode (preset_param/kerastuner)")
parser.add_argument("--batch_size",
type=int,
default=64,
help="Batch size")
parser.add_argument("--sen_len",
type=int,
default=20,
help="Maximum length of a sentence")
parser.add_argument("--lstm1",
type=int,
default=32,
help="Hidden dimension of first LSTM")
parser.add_argument("--lstm2",
type=int,
default=32,
help="Hidden dimension of second LSTM")
parser.add_argument("--dp_rate",
type=float,
default=0.5,
help="Dropout rate (percentage of droping)")
parser.add_argument("--lr", type=float, default=1e-3, help="Learning rate")
parser.add_argument("--epochs", type=int, default=1, help="epochs")
parser.add_argument("--epochs_before_search",
type=int,
default=1,
help="epochs_before_search")
parser.add_argument("--epochs_after_search",
type=int,
default=1,
help="epochs_after_search")
parser.add_argument("--max_trials",
type=int,
default=1,
help="max_trials for kerastuner")
parser.add_argument("--executions_per_trial",
type=int,
default=1,
help="executions_per_trial for kerastuner")
args = parser.parse_args()
# Setup paths
path_prefix = Path.cwd()
train_with_label = os.path.join(path_prefix, 'data/training_label.txt')
train_no_label = os.path.join(path_prefix, 'data/training_nolabel.txt')
testing_data = os.path.join(path_prefix, 'data/testing_data.txt')
w2v_path = path_prefix.joinpath('model/w2v_all.model')
# Configuration
batch_size = args.batch_size
sen_len = args.sen_len
# Preprocess dataset
## Read 'training_label.txt' and 'training_nolabel.txt'
print("loading training data ...")
X_train_lable, y_train_lable = load_training_data(train_with_label)
X_train, X_val, y_train, y_val = train_test_split(X_train_lable,
y_train_lable,
test_size=0.1)
train_x_no_label = load_training_data(train_no_label)
print(
f"Positive rate in training dataset: {np.sum(y_train) / len(y_train)}")
print(f"Positive rate in validation dataset: {np.sum(y_val) / len(y_val)}")
## Build the preprocessor
preprocessor = Preprocess(sen_len, w2v_path=str(w2v_path))
embedding = preprocessor.make_embedding(load=True)
X_train_idx = preprocessor.sentences_word2idx(X_train)
X_val_idx = preprocessor.sentences_word2idx(X_val)
print(f"Pretrained embedding matrix shape: {embedding.shape}")
## Preprocess training and validation datasets
X_train_idx_dataset = tf.data.Dataset.from_tensor_slices(X_train_idx)
y_train_dataset = tf.data.Dataset.from_tensor_slices(y_train)
train_dataset = tf.data.Dataset.zip((X_train_idx_dataset, y_train_dataset))
X_val_idx_dataset = tf.data.Dataset.from_tensor_slices(X_val_idx)
y_val_dataset = tf.data.Dataset.from_tensor_slices(y_val)
val_dataset = tf.data.Dataset.zip((X_val_idx_dataset, y_val_dataset))
train_dataset = train_dataset.batch(batch_size)
val_dataset = val_dataset.batch(batch_size)
train_dataset = train_dataset.cache().prefetch(AUTOTUNE)
val_dataset = val_dataset.cache().prefetch(AUTOTUNE)
# Train a bidirectional LSTM model
train_embedding = False # fix embedding during training
## Method1 - preset parameters
if args.train_mode == 'preset_param':
### Build the model
hidden_dim1 = args.lstm1
hidden_dim2 = args.lstm2
dp_rate = args.dp_rate
lr = args.lr
epochs = args.epochs
model = buildModel(embedding, train_embedding, sen_len, hidden_dim1,
hidden_dim2, dp_rate, lr)
model.summary()
### Train the model
checkpoint_filepath = os.path.join(path_prefix, 'ckpt/')
model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=checkpoint_filepath, save_best_only=True)
history = model.fit(train_dataset,
validation_data=val_dataset,
epochs=epochs,
callbacks=[model_checkpoint_callback])
elif args.train_mode == 'kerastuner':
import IPython
from kerastuner.tuners import RandomSearch
class ClearTrainingOutput(tf.keras.callbacks.Callback):
def on_train_end(*args, **kwargs):
IPython.display.clear_output(wait=True)
### Build the model
tuner = RandomSearch(BiLstmTuner(embedding, train_embedding, sen_len),
objective='val_accuracy',
max_trials=args.max_trials,
executions_per_trial=args.executions_per_trial,
directory=os.path.join(path_prefix, 'tuner_dir'),
project_name='tsc')
### Train the model
tuner.search(
train_dataset,
epochs=args.epochs_before_search,
validation_data=val_dataset,
verbose=1,
callbacks=[ClearTrainingOutput()],
)
# Load the best model
print('\nload model ...')
## Method1
if args.train_mode == 'preset_param':
best_model = tf.keras.models.load_model(checkpoint_filepath)
## Method2
elif args.train_mode == 'kerastuner':
tuner.results_summary(num_trials=min(3, args.max_trials))
best_model = tuner.get_best_models()[0]
best_model.summary()
# Train again with training set and validation set
combined_dataset = train_dataset.concatenate(val_dataset)
best_model.fit(combined_dataset, epochs=args.epochs_after_search)
# Testing
## Preprocess test dataset
print("loading testing data ...")
X_test = load_testing_data(testing_data)
X_test_idx = preprocessor.sentences_word2idx(X_test)
test_dataset = tf.data.Dataset.from_tensor_slices(X_test_idx)
test_dataset = test_dataset.batch(batch_size)
test_dataset = test_dataset.cache().prefetch(AUTOTUNE)
## Predict
outputs = testing(best_model, test_dataset)
# Write the result to a CSV file
tmp = pd.DataFrame({
"id": [str(i) for i in range(len(X_test))],
"label": outputs
})
print("save csv ...")
tmp.to_csv(os.path.join(path_prefix, 'predict.csv'), index=False)
print("Finish Predicting")
def _run():
rand = np.random.RandomState(42)
layers, weights = _model.buildModel(rand, batchSize=None, sampleMode=4)
for k, v in weights.items():
print("Weight '{}' has shape {}".format(k, v.get_shape()))
for x in layers:
print("Layer '{}' has shape {}".format(x.name, x.get_shape()))
g_x = layers[0]
g_logits = layers[-1]
shapeDst = tuple(g_logits.get_shape().as_list())
g_y = tf.placeholder(tf.float32, (None, ))
g_yRep = tf.tile(tf.reshape(g_y, shape=(-1, 1, 1, 1)),
multiples=(1, ) + shapeDst[1:])
# Learning rate placeholder.
g_lr = tf.placeholder(tf.float32, ())
# For reporting statistics.
g_correct = tf.equal(tf.cast(tf.greater(g_logits, 0.0), tf.float32),
g_yRep)
g_correct_sum = tf.reduce_sum(tf.cast(
g_correct, tf.int32)) / (shapeDst[1] * shapeDst[2])
# Apply sigmoid and cross entropy.
g_ce = tf.nn.sigmoid_cross_entropy_with_logits(labels=g_yRep,
logits=g_logits)
# We'll use the loss-sum when reporting statistics, and the loss-mean when training.
g_loss_sum = tf.reduce_sum(g_ce)
g_loss_mean = tf.reduce_mean(g_ce)
# Use Adam with the place-holder learning rate.
g_optimizer = tf.train.AdamOptimizer(learning_rate=g_lr)
g_train_step = g_optimizer.minimize(g_loss_mean)
def _eval(sess, xs, ys, batchSize=64):
""" Returns the number of correct predictions, sum of losses, and the number of samples.
The caller can divide the first two by the third to get averages.
"""
assert batchSize > 0
assert xs.shape[0] == ys.shape[0]
num = xs.shape[0]
acc_sum = 0
loss_sum = 0.0
for ii in range(0, num, batchSize):
xsCur = xs[ii:ii + batchSize]
ysCur = ys[ii:ii + batchSize]
acc, loss = sess.run((g_correct_sum, g_loss_sum),
feed_dict={
g_x: xsCur,
g_y: ysCur
})
acc_sum += acc
loss_sum += loss
return acc_sum, loss_sum, num
def _trainOne(sess, xs, ys, rate, batchSize=64):
""" Performs one training epoch. """
assert batchSize > 0
assert xs.shape[0] == ys.shape[0]
num = xs.shape[0]
# Generate a random permutation of the training data.
indices = rand.permutation(num)
# Run the optimizer on each batch.
for ii in range(0, num, batchSize):
inds = indices[ii:ii + batchSize]
xsCur = xs[inds]
ysCur = ys[inds]
sess.run(g_train_step,
feed_dict={
g_x: xsCur,
g_y: ysCur,
g_lr: rate
})
def _trainMulti(sess,
epochs,
xsTrain,
ysTrain,
xsValid,
ysValid,
rate,
batchSize=64):
""" Performs multiple epochs, and reports statistics after each. """
assert epochs > 0
print("*** Starting {} epochs with batch size {} and learning rate {}".
format(epochs, batchSize, rate))
for i in range(epochs):
t0 = time.time()
_trainOne(sess, xsTrain, ysTrain, rate, batchSize)
t1 = time.time()
print(" Epoch {} took: {:.04f} sec".format(i, t1 - t0))
acc, loss, num = _eval(sess, xsTrain, ysTrain, batchSize)
print(" Training accuracy and loss: {:.03f}, {:.03f}".format(
acc / num, loss / num))
acc, loss, num = _eval(sess, xsValid, ysValid, batchSize)
print(" Validation accuracy and loss: {:.03f}, {:.03f}".format(
acc / num, loss / num))
featuresTrain, labelsTrain, featuresValid, labelsValid = _loadAndSplitData(
rand, frac=0.9)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
# Report the initial statistics (before any epochs). Generally, the accuracy is expected to be around 1 / 2.
acc, loss, num = _eval(sess, featuresTrain, labelsTrain)
print("Training accuracy and loss before: {:.03f}, {:.03f}".format(
acc / num, loss / num))
acc, loss, num = _eval(sess, featuresValid, labelsValid)
print("Validation accuracy and loss before: {:.03f}, {:.03f}".format(
acc / num, loss / num))
batchSize = 64
_trainMulti(sess,
5,
featuresTrain,
labelsTrain,
featuresValid,
labelsValid,
rate=0.00100,
batchSize=batchSize)
_trainMulti(sess,
3,
featuresTrain,
labelsTrain,
featuresValid,
labelsValid,
rate=0.00030,
batchSize=batchSize)
_trainMulti(sess,
3,
featuresTrain,
labelsTrain,
featuresValid,
labelsValid,
rate=0.00010,
batchSize=batchSize)
# _trainMulti(sess, 5, featuresTrain, labelsTrain, featuresValid, labelsValid, rate=0.00005, batchSize=batchSize)
_model.saveModelWeights(sess, weights)