def __init__(self, image_nr, image_resize, model_name):
super().__init__()
self.model_name = model_name
self.ctx = neptune.Context()
self.epoch_loss_averager = Averager()
self.image_nr = image_nr
self.image_resize = image_resize
def __init__(self, min_lr=1e-8, multipy_factor=1.05, add_factor=0.0):
super().__init__()
self.ctx = neptune.Context()
self.ctx.channel_reset('Learning Rate Finder')
self.min_lr = min_lr
self.multipy_factor = multipy_factor
self.add_factor = add_factor
def __init__(self, metric_name, minimize, reduce_factor, reduce_patience, min_lr):
super().__init__()
self.ctx = neptune.Context()
self.ctx.channel_reset('Learning Rate')
self.metric_name = metric_name
self.minimize = minimize
self.reduce_factor = reduce_factor
self.reduce_patience = reduce_patience
self.min_lr = min_lr
def __init__(self, image_nr, image_resize, image_every, model_name,
use_depth):
super().__init__()
self.model_name = model_name
self.ctx = neptune.Context()
self.epoch_loss_averager = Averager()
self.image_resize = image_resize
self.image_every = image_every
self.image_nr = image_nr
self.use_depth = use_depth
def __init__(self, number_of_batches_per_full_cycle, max_lr, enabled=1, momentum_range=(0.95, 0.8),
prcnt_annihilate=10,
div=10):
super().__init__()
self.enabled = enabled
self.number_of_batches_per_full_cycle = number_of_batches_per_full_cycle
self.max_lr = max_lr
self.momentum_range = momentum_range
self.prcnt_annihilate = prcnt_annihilate
self.div = div
self.ctx = neptune.Context()
def main():
"""
Load data and train a model on it.
"""
context = neptune.Context()
context.integrate_with_tensorflow()
final_train_channel = context.create_channel('final_train_accuracy',
neptune.ChannelType.NUMERIC)
final_test_channel = context.create_channel('final_test_accuracy',
neptune.ChannelType.NUMERIC)
args = neptune_args(context)
print('args:\n', args)
random.seed(args.seed)
train_set, test_set = split_dataset(read_dataset(args.omniglot_src))
train_set = list(augment_dataset(train_set))
test_set = list(test_set)
model = ProgressiveOmniglotModel(args.classes, **model_kwargs(args))
config = tf.ConfigProto()
config.gpu_options.allow_growth = args.allow_growth
with tf.Session(config=config) as sess:
if not args.pretrained:
print('Training...')
train(sess, model, train_set, test_set, args.checkpoint,
**train_kwargs(args))
else:
print('Restoring from checkpoint...')
tf.train.Saver().restore(
sess, tf.train.latest_checkpoint(args.checkpoint))
print('Evaluating...')
eval_kwargs = evaluate_kwargs(args)
final_train_accuracy = evaluate(sess, model, train_set, **eval_kwargs)
print('final_train_accuracy:', final_train_accuracy)
final_train_channel.send(final_train_accuracy)
final_test_accuracy = evaluate(sess, model, test_set, **eval_kwargs)
print('final_test_accuracy:', final_test_accuracy)
final_test_channel.send(final_test_accuracy)
from attrdict import AttrDict import neptune import numpy as np import pandas as pd from sklearn.externals import joblib from steppy.base import Step, IdentityOperation from steppy.adapter import Adapter, E from common_blocks import augmentation as aug from common_blocks import metrics from common_blocks import models from common_blocks import pipelines from common_blocks import utils from common_blocks import postprocessing CTX = neptune.Context() LOGGER = utils.init_logger() # ______ ______ .__ __. _______ __ _______ _______. # / | / __ \ | \ | | | ____|| | / _____| / | # | ,----'| | | | | \| | | |__ | | | | __ | (----` # | | | | | | | . ` | | __| | | | | |_ | \ \ # | `----.| `--' | | |\ | | | | | | |__| | .----) | # \______| \______/ |__| \__| |__| |__| \______| |_______/ # EXPERIMENT_DIR = '/output/experiment' CLONE_EXPERIMENT_DIR_FROM = '' # When running eval in the cloud specify this as for example /input/SAL-14/output/experiment OVERWRITE_EXPERIMENT_DIR = False DEV_MODE = False
def __init__(self):
super().__init__()
self.ctx = neptune.Context()
self.best_loss = None
import neptune
ctx = neptune.Context()
def neptune_send_plot(logs):
epoch_data = logs[-1]
for metrics, value in epoch_data.items():
ctx.channel_send(name=metrics, y=value)
