import keras
import numpy as np
import os
import pandas as pd
import pescador
import tensorflow as tf
import time
import oyaml as yaml
import localmodule
from models import create_flat_singletask_coarse_model
from keras.optimizers import Adam
# Define constants.
train_data_dir = localmodule.get_train_data_dir()
train_dataset_name = localmodule.get_train_dataset_name()
valid_data_dir = localmodule.get_valid_data_dir()
valid_dataset_name = localmodule.get_valid_dataset_name()
models_dir = localmodule.get_models_dir()
n_input_hops = 104
n_filters = [24, 48, 48]
kernel_size = [5, 5]
pool_size = [2, 4]
n_hidden_units = 64
# Read command-line arguments.
parser = argparse.ArgumentParser()
parser.add_argument('aug_kind_str')
parser.add_argument('trial_str')
parser.add_argument('--lr', type=float, default=1e-4)
parser.add_argument('--base-wd', type=float, default=1e-4)
parser.add_argument('--batch-size', type=int, default=128)
min_valid_idxs = valid_idxs
min_valid_distr = valid_distr
min_train_distr = train_distr
# Get list of train and validation "sources"
train_sources = []
valid_sources = []
for idx, source in enumerate(source_name_list):
if idx in min_valid_idxs:
valid_sources.append(source)
else:
train_sources.append(source)
# Copy train and validation data
train_dir = os.path.join(localmodule.get_train_data_dir(), "original")
valid_dir = os.path.join(localmodule.get_valid_data_dir(), "original")
for root, dirs, files in os.walk(anafcc_dir):
for fname in files:
for source in train_sources:
if (fname.endswith('.hdf5')
or fname.endswith('.h5')) and fname.startswith(source):
src_path = os.path.join(root, fname)
rel_path = src_path.replace(anafcc_dir, '')
dst_path = os.path.join(train_dir, rel_path)
os.makedirs(os.path.dirname(dst_path), exist_ok=True)
shutil.copy(src_path, dst_path)
for source in valid_sources:
if (fname.endswith('.hdf5')
or fname.endswith('.h5')) and fname.startswith(source):
src_path = os.path.join(root, fname)
rel_path = src_path.replace(anafcc_dir, '')