snapshot...

modules/deeplearning/cloud_opd_srcnn.py

+import glob
+import tensorflow as tf
+
+import util.util
+from util.setup import logdir, modeldir, now, ancillary_path
+from util.util import EarlyStop, normalize, denormalize, get_grid_values_all, resample_2d_linear
+import os, datetime
+import numpy as np
+import pickle
+import h5py
+
+LOG_DEVICE_PLACEMENT = False
+
+PROC_BATCH_SIZE = 4
+PROC_BATCH_BUFFER_SIZE = 5000
+
+NumClasses = 5
+if NumClasses == 2:
+    NumLogits = 1
+else:
+    NumLogits = NumClasses
+
+BATCH_SIZE = 128
+NUM_EPOCHS = 80
+
+TRACK_MOVING_AVERAGE = False
+EARLY_STOP = True
+
+NOISE_TRAINING = False
+NOISE_STDDEV = 0.01
+DO_AUGMENT = True
+
+DO_SMOOTH = False
+SIGMA = 1.0
+DO_ZERO_OUT = False
+
+# setup scaling parameters dictionary
+mean_std_dct = {}
+mean_std_file = ancillary_path+'mean_std_lo_hi_l2.pkl'
+f = open(mean_std_file, 'rb')
+mean_std_dct_l2 = pickle.load(f)
+f.close()
+
+mean_std_file = ancillary_path+'mean_std_lo_hi_l1b.pkl'
+f = open(mean_std_file, 'rb')
+mean_std_dct_l1b = pickle.load(f)
+f.close()
+
+mean_std_dct.update(mean_std_dct_l1b)
+mean_std_dct.update(mean_std_dct_l2)
+
+IMG_DEPTH = 1
+
+label_param = 'cld_opd_dcomp'
+
+params = ['temp_11_0um_nom', 'refl_0_65um_nom', label_param]
+params_i = ['refl_0_65um_nom', label_param]
+data_params_half = ['temp_11_0um_nom']
+data_params_full = ['refl_0_65um_nom']
+
+label_idx_i = params_i.index(label_param)
+label_idx = params.index(label_param)
+
+print('data_params_half: ', data_params_half)
+print('data_params_full: ', data_params_full)
+print('label_param: ', label_param)
+
+KERNEL_SIZE = 3  # target size: (128, 128)
+N_X = N_Y = 1
+
+if KERNEL_SIZE == 3:
+    slc_x = slice(2, N_X*128 + 4)
+    slc_y = slice(2, N_Y*128 + 4)
+    slc_x_2 = slice(1, N_X*128 + 6, 2)
+    slc_y_2 = slice(1, N_Y*128 + 6, 2)
+    x_2 = np.arange(int((N_X*128)/2) + 3)
+    y_2 = np.arange(int((N_Y*128)/2) + 3)
+    t = np.arange(0, int((N_X*128)/2) + 3, 0.5)
+    s = np.arange(0, int((N_Y*128)/2) + 3, 0.5)
+    x_k = slice(1, N_X*128 + 3)
+    y_k = slice(1, N_Y*128 + 3)
+    x_128 = slice(3, N_X*128 + 3)
+    y_128 = slice(3, N_Y*128 + 3)
+elif KERNEL_SIZE == 5:
+    slc_x = slice(3, 135)
+    slc_y = slice(3, 135)
+    slc_x_2 = slice(2, 137, 2)
+    slc_y_2 = slice(2, 137, 2)
+    x_128 = slice(5, 133)
+    y_128 = slice(5, 133)
+    t = np.arange(1, 67, 0.5)
+    s = np.arange(1, 67, 0.5)
+    x_2 = np.arange(68)
+    y_2 = np.arange(68)
+# ----------------------------------------
+
+
+def build_residual_conv2d_block(conv, num_filters, block_name, activation=tf.nn.relu, padding='SAME',
+                                kernel_initializer='he_uniform', scale=None, kernel_size=3,
+                                do_drop_out=True, drop_rate=0.5, do_batch_norm=True):
+
+    with tf.name_scope(block_name):
+        skip = tf.keras.layers.Conv2D(num_filters, kernel_size=kernel_size, padding=padding, kernel_initializer=kernel_initializer, activation=activation)(conv)
+        skip = tf.keras.layers.Conv2D(num_filters, kernel_size=kernel_size, padding=padding, activation=None)(skip)
+
+        if scale is not None:
+            skip = tf.keras.layers.Lambda(lambda x: x * scale)(skip)
+
+        if do_drop_out:
+            skip = tf.keras.layers.Dropout(drop_rate)(skip)
+
+        if do_batch_norm:
+            skip = tf.keras.layers.BatchNormalization()(skip)
+
+        conv = conv + skip
+        print(block_name+':', conv.shape)
+
+    return conv
+
+
+def upsample(tmp):
+    tmp = tmp[:, slc_y_2, slc_x_2]
+    tmp = resample_2d_linear(x_2, y_2, tmp, t, s)
+    tmp = tmp[:, y_k, x_k]
+    return tmp
+
+
+def upsample_nearest(grd):
+    bsize, ylen, xlen = grd.shape
+    up = np.zeros((bsize, ylen*2, xlen*2))
+
+    up[:, 0::2, 0::2] = grd[:, 0::, 0::]
+    up[:, 1::2, 0::2] = grd[:, 0::, 0::]
+    up[:, 0::2, 1::2] = grd[:, 0::, 0::]
+    up[:, 1::2, 1::2] = grd[:, 0::, 0::]
+
+    return up
+
+
+def upsample_mean(grd):
+    bsize, ylen, xlen = grd.shape
+    up = np.zeros((bsize, ylen*2, xlen*2))
+
+    up[:, ::2, ::2] = grd[:, ::2, ::2]
+    up[:, 1::2, ::2] = grd[:, ::2, ::2]
+    up[:, ::2, 1::2] = grd[:, ::2, ::2]
+    up[:, 1::2, 1::2] = grd[:, ::2, ::2]
+
+    return up
+
+
+def get_grid_cell_mean(grd_k):
+    grd_k = np.where(np.isnan(grd_k), 0, grd_k)
+    a = grd_k[:, 0::2, 0::2]
+    b = grd_k[:, 1::2, 0::2]
+    c = grd_k[:, 0::2, 1::2]
+    d = grd_k[:, 1::2, 1::2]
+    mean = np.nanmean([a, b, c, d], axis=0)
+
+    return mean
+
+
+def get_min_max_std(grd_k):
+    grd_k = np.where(np.isnan(grd_k), 0, grd_k)
+    a = grd_k[:, 0::2, 0::2]
+    b = grd_k[:, 1::2, 0::2]
+    c = grd_k[:, 0::2, 1::2]
+    d = grd_k[:, 1::2, 1::2]
+
+    lo = np.nanmin([a, b, c, d], axis=0)
+    hi = np.nanmax([a, b, c, d], axis=0)
+    std = np.nanstd([a, b, c, d], axis=0)
+    avg = np.nanmean([a, b, c, d], axis=0)
+
+    return lo, hi, std, avg
+
+
+class SRCNN:
+    
+    def __init__(self):
+
+        self.train_data = None
+        self.train_label = None
+        self.test_data = None
+        self.test_label = None
+        self.test_data_denorm = None
+        
+        self.train_dataset = None
+        self.inner_train_dataset = None
+        self.test_dataset = None
+        self.eval_dataset = None
+        self.X_img = None
+        self.X_prof = None
+        self.X_u = None
+        self.X_v = None
+        self.X_sfc = None
+        self.inputs = []
+        self.y = None
+        self.handle = None
+        self.inner_handle = None
+        self.in_mem_batch = None
+
+        self.h5f_l1b_trn = None
+        self.h5f_l1b_tst = None
+        self.h5f_l2_trn = None
+        self.h5f_l2_tst = None
+
+        self.logits = None
+
+        self.predict_data = None
+        self.predict_dataset = None
+        self.mean_list = None
+        self.std_list = None
+        
+        self.training_op = None
+        self.correct = None
+        self.accuracy = None
+        self.loss = None
+        self.pred_class = None
+        self.variable_averages = None
+
+        self.global_step = None
+
+        self.writer_train = None
+        self.writer_valid = None
+        self.writer_train_valid_loss = None
+
+        self.OUT_OF_RANGE = False
+
+        self.model = None
+        self.optimizer = None
+        self.ema = None
+        self.train_loss = None
+        self.train_accuracy = None
+        self.test_loss = None
+        self.test_accuracy = None
+        self.test_auc = None
+        self.test_recall = None
+        self.test_precision = None
+        self.test_confusion_matrix = None
+        self.test_true_pos = None
+        self.test_true_neg = None
+        self.test_false_pos = None
+        self.test_false_neg = None
+
+        self.test_labels = []
+        self.test_preds = []
+        self.test_probs = None
+
+        self.learningRateSchedule = None
+        self.num_data_samples = None
+        self.initial_learning_rate = None
+
+        self.data_dct = None
+        self.train_data_files = None
+        self.train_label_files = None
+        self.test_data_files = None
+        self.test_label_files = None
+
+        # self.n_chans = len(data_params_half) + len(data_params_full) + 1
+        self.n_chans = 5
+
+        self.X_img = tf.keras.Input(shape=(None, None, self.n_chans))
+
+        self.inputs.append(self.X_img)
+
+        tf.debugging.set_log_device_placement(LOG_DEVICE_PLACEMENT)
+
+    def get_in_mem_data_batch(self, idxs, is_training):
+        if is_training:
+            data_files = self.train_data_files
+            label_files = self.train_label_files
+        else:
+            data_files = self.test_data_files
+            label_files = self.test_label_files
+
+        data_s = []
+        label_s = []
+        for k in idxs:
+            f = data_files[k]
+            nda = np.load(f)
+            data_s.append(nda)
+
+            f = label_files[k]
+            nda = np.load(f)
+            label_s.append(nda)
+        input_data = np.concatenate(data_s)
+        input_label = np.concatenate(label_s)
+
+        data_norm = []
+        for param in data_params_half:
+            idx = params.index(param)
+            tmp = input_data[:, idx, :, :]
+            tmp = np.where(np.isnan(tmp), 0, tmp)
+            tmp = upsample(tmp)
+            tmp = normalize(tmp, param, mean_std_dct)
+            data_norm.append(tmp)
+
+        for param in data_params_full:
+            idx = params_i.index(param)
+            tmp = input_label[:, idx, :, :]
+            tmp = np.where(np.isnan(tmp), 0, tmp)
+
+            lo, hi, std, avg = get_min_max_std(tmp)
+            lo = upsample_nearest(lo)
+            hi = upsample_nearest(hi)
+            avg = upsample_nearest(avg)
+            lo = normalize(lo, param, mean_std_dct)
+            hi = normalize(hi, param, mean_std_dct)
+            avg = normalize(avg, param, mean_std_dct)
+
+            data_norm.append(lo[:, slc_y, slc_x])
+            data_norm.append(hi[:, slc_y, slc_x])
+            data_norm.append(avg[:, slc_y, slc_x])
+        # ---------------------------------------------------
+        tmp = input_data[:, label_idx, :, :]
+        tmp = np.where(np.isnan(tmp), 0, tmp)
+        tmp = upsample(tmp)
+        data_norm.append(tmp)
+        # ---------
+        data = np.stack(data_norm, axis=3)
+        data = data.astype(np.float32)
+
+        # -----------------------------------------------------
+        # -----------------------------------------------------
+        label = input_label[:, label_idx_i, :, :]
+        label = label.copy()
+        label = label[:, y_128, x_128]
+
+        label = np.where(np.isnan(label), 0, label)
+        label = np.expand_dims(label, axis=3)
+
+        data = data.astype(np.float32)
+        label = label.astype(np.float32)
+
+        if is_training and DO_AUGMENT:
+            data_ud = np.flip(data, axis=1)
+            label_ud = np.flip(label, axis=1)
+
+            data_lr = np.flip(data, axis=2)
+            label_lr = np.flip(label, axis=2)
+
+            data = np.concatenate([data, data_ud, data_lr])
+            label = np.concatenate([label, label_ud, label_lr])
+
+        return data, label
+
+    def get_in_mem_data_batch_train(self, idxs):
+        return self.get_in_mem_data_batch(idxs, True)
+
+    def get_in_mem_data_batch_test(self, idxs):
+        return self.get_in_mem_data_batch(idxs, False)
+
+    @tf.function(input_signature=[tf.TensorSpec(None, tf.int32)])
+    def data_function(self, indexes):
+        out = tf.numpy_function(self.get_in_mem_data_batch_train, [indexes], [tf.float32, tf.float32])
+        return out
+
+    @tf.function(input_signature=[tf.TensorSpec(None, tf.int32)])
+    def data_function_test(self, indexes):
+        out = tf.numpy_function(self.get_in_mem_data_batch_test, [indexes], [tf.float32, tf.float32])
+        return out
+
+    def get_train_dataset(self, indexes):
+        indexes = list(indexes)
+
+        dataset = tf.data.Dataset.from_tensor_slices(indexes)
+        dataset = dataset.batch(PROC_BATCH_SIZE)
+        dataset = dataset.map(self.data_function, num_parallel_calls=8)
+        dataset = dataset.cache()
+        if DO_AUGMENT:
+            dataset = dataset.shuffle(PROC_BATCH_BUFFER_SIZE)
+        dataset = dataset.prefetch(buffer_size=1)
+        self.train_dataset = dataset
+
+    def get_test_dataset(self, indexes):
+        indexes = list(indexes)
+
+        dataset = tf.data.Dataset.from_tensor_slices(indexes)
+        dataset = dataset.batch(PROC_BATCH_SIZE)
+        dataset = dataset.map(self.data_function_test, num_parallel_calls=8)
+        dataset = dataset.cache()
+        self.test_dataset = dataset
+
+    def setup_pipeline(self, train_data_files, train_label_files, test_data_files, test_label_files, num_train_samples):
+        self.train_data_files = train_data_files
+        self.train_label_files = train_label_files
+        self.test_data_files = test_data_files
+        self.test_label_files = test_label_files
+
+        trn_idxs = np.arange(len(train_data_files))
+        np.random.shuffle(trn_idxs)
+
+        tst_idxs = np.arange(len(test_data_files))
+
+        self.get_train_dataset(trn_idxs)
+        self.get_test_dataset(tst_idxs)
+
+        self.num_data_samples = num_train_samples  # approximately
+
+        print('datetime: ', now)
+        print('training and test data: ')
+        print('---------------------------')
+        print('num train samples: ', self.num_data_samples)
+        print('BATCH SIZE: ', BATCH_SIZE)
+        print('num test samples: ', tst_idxs.shape[0])
+        print('setup_pipeline: Done')
+
+    def setup_test_pipeline(self, test_data_files, test_label_files):
+        self.test_data_files = test_data_files
+        self.test_label_files = test_label_files
+        tst_idxs = np.arange(len(test_data_files))
+        self.get_test_dataset(tst_idxs)
+        print('setup_test_pipeline: Done')
+
+    def build_srcnn(self, do_drop_out=False, do_batch_norm=False, drop_rate=0.5, factor=2):
+        print('build_cnn')
+        padding = "SAME"
+
+        # activation = tf.nn.relu
+        # activation = tf.nn.elu
+        activation = tf.nn.relu
+        momentum = 0.99
+
+        num_filters = 64
+
+        input_2d = self.inputs[0]
+        print('input: ', input_2d.shape)
+
+        conv = conv_b = tf.keras.layers.Conv2D(num_filters, kernel_size=KERNEL_SIZE, kernel_initializer='he_uniform', activation=activation, padding='VALID')(input_2d)
+        print(conv.shape)
+
+        # if NOISE_TRAINING:
+        #     conv = conv_b = tf.keras.layers.GaussianNoise(stddev=NOISE_STDDEV)(conv)
+
+        scale = 0.2
+
+        conv_b = build_residual_conv2d_block(conv_b, num_filters, 'Residual_Block_1', kernel_size=KERNEL_SIZE, scale=scale)
+
+        conv_b = build_residual_conv2d_block(conv_b, num_filters, 'Residual_Block_2', kernel_size=KERNEL_SIZE, scale=scale)
+
+        conv_b = build_residual_conv2d_block(conv_b, num_filters, 'Residual_Block_3', kernel_size=KERNEL_SIZE, scale=scale)
+
+        conv_b = build_residual_conv2d_block(conv_b, num_filters, 'Residual_Block_4', kernel_size=KERNEL_SIZE, scale=scale)
+
+        conv_b = build_residual_conv2d_block(conv_b, num_filters, 'Residual_Block_5', kernel_size=KERNEL_SIZE, scale=scale)
+
+        conv_b = build_residual_conv2d_block(conv_b, num_filters, 'Residual_Block_6', kernel_size=KERNEL_SIZE, scale=scale)
+
+        conv_b = tf.keras.layers.Conv2D(num_filters, kernel_size=3, strides=1, activation=activation, kernel_initializer='he_uniform', padding=padding)(conv_b)
+
+        conv = conv + conv_b
+        print(conv.shape)
+
+        # This is effectively a Dense layer
+        self.logits = tf.keras.layers.Conv2D(1, kernel_size=1, strides=1, padding=padding, name='regression')(conv)
+        print(self.logits.shape)
+
+    def build_training(self):
+
+        self.loss = tf.keras.losses.MeanSquaredError()  # Regression
+
+        # decayed_learning_rate = learning_rate * decay_rate ^ (global_step / decay_steps)
+        initial_learning_rate = 0.002
+        decay_rate = 0.95
+        steps_per_epoch = int(self.num_data_samples/BATCH_SIZE)  # one epoch
+        decay_steps = int(steps_per_epoch) * 2
+        print('initial rate, decay rate, steps/epoch, decay steps: ', initial_learning_rate, decay_rate, steps_per_epoch, decay_steps)
+
+        self.learningRateSchedule = tf.keras.optimizers.schedules.ExponentialDecay(initial_learning_rate, decay_steps, decay_rate)
+
+        optimizer = tf.keras.optimizers.Adam(learning_rate=self.learningRateSchedule)
+
+        if TRACK_MOVING_AVERAGE:
+            # Not really sure this works properly (from tfa)
+            # optimizer = tfa.optimizers.MovingAverage(optimizer)
+            self.ema = tf.train.ExponentialMovingAverage(decay=0.9999)
+
+        self.optimizer = optimizer
+        self.initial_learning_rate = initial_learning_rate
+
+    def build_evaluation(self):
+        self.train_loss = tf.keras.metrics.Mean(name='train_loss')
+        self.test_loss = tf.keras.metrics.Mean(name='test_loss')
+
+        if NumClasses == 2:
+            self.train_accuracy = tf.keras.metrics.BinaryAccuracy(name='train_accuracy')
+            self.test_accuracy = tf.keras.metrics.BinaryAccuracy(name='test_accuracy')
+            self.test_auc = tf.keras.metrics.AUC(name='test_auc')
+            self.test_recall = tf.keras.metrics.Recall(name='test_recall')
+            self.test_precision = tf.keras.metrics.Precision(name='test_precision')
+            self.test_true_neg = tf.keras.metrics.TrueNegatives(name='test_true_neg')
+            self.test_true_pos = tf.keras.metrics.TruePositives(name='test_true_pos')
+            self.test_false_neg = tf.keras.metrics.FalseNegatives(name='test_false_neg')
+            self.test_false_pos = tf.keras.metrics.FalsePositives(name='test_false_pos')
+        else:
+            self.train_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='train_accuracy')
+            self.test_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='test_accuracy')
+
+    @tf.function(input_signature=[tf.TensorSpec(None, tf.float32), tf.TensorSpec(None, tf.float32)])
+    def train_step(self, inputs, labels):
+        labels = tf.squeeze(labels, axis=[3])
+        with tf.GradientTape() as tape:
+            pred = self.model([inputs], training=True)
+            loss = self.loss(labels, pred)
+            total_loss = loss
+            if len(self.model.losses) > 0:
+                reg_loss = tf.math.add_n(self.model.losses)
+                total_loss = loss + reg_loss
+        gradients = tape.gradient(total_loss, self.model.trainable_variables)
+        self.optimizer.apply_gradients(zip(gradients, self.model.trainable_variables))
+        if TRACK_MOVING_AVERAGE:
+            self.ema.apply(self.model.trainable_variables)
+
+        self.train_loss(loss)
+        self.train_accuracy(labels, pred)
+
+        return loss
+
+    @tf.function(input_signature=[tf.TensorSpec(None, tf.float32), tf.TensorSpec(None, tf.float32)])
+    def test_step(self, inputs, labels):
+        labels = tf.squeeze(labels, axis=[3])
+        pred = self.model([inputs], training=False)
+        t_loss = self.loss(labels, pred)
+
+        self.test_loss(t_loss)
+        self.test_accuracy(labels, pred)
+
+    # @tf.function(input_signature=[tf.TensorSpec(None, tf.float32), tf.TensorSpec(None, tf.float32)])
+    # decorator commented out because pred.numpy(): pred not evaluated yet.
+    def predict(self, inputs, labels):
+        pred = self.model([inputs], training=False)
+        # t_loss = self.loss(tf.squeeze(labels, axis=[3]), pred)
+        t_loss = self.loss(labels, pred)
+
+        self.test_labels.append(labels)
+        self.test_preds.append(pred.numpy())
+
+        self.test_loss(t_loss)
+        self.test_accuracy(labels, pred)
+
+    def reset_test_metrics(self):
+        self.test_loss.reset_states()
+        self.test_accuracy.reset_states()
+
+    def get_metrics(self):
+        recall = self.test_recall.result()
+        precsn = self.test_precision.result()
+        f1 = 2 * (precsn * recall) / (precsn + recall)
+
+        tn = self.test_true_neg.result()
+        tp = self.test_true_pos.result()
+        fn = self.test_false_neg.result()
+        fp = self.test_false_pos.result()
+
+        mcc = ((tp * tn) - (fp * fn)) / np.sqrt((tp + fp) * (tp + fn) * (tn + fp) * (tn + fn))
+        return f1, mcc
+
+    def do_training(self, ckpt_dir=None):
+
+        if ckpt_dir is None:
+            if not os.path.exists(modeldir):
+                os.mkdir(modeldir)
+            ckpt = tf.train.Checkpoint(step=tf.Variable(1), model=self.model)
+            ckpt_manager = tf.train.CheckpointManager(ckpt, modeldir, max_to_keep=3)
+        else:
+            ckpt = tf.train.Checkpoint(step=tf.Variable(1), model=self.model)
+            ckpt_manager = tf.train.CheckpointManager(ckpt, ckpt_dir, max_to_keep=3)
+            ckpt.restore(ckpt_manager.latest_checkpoint)
+
+        self.writer_train = tf.summary.create_file_writer(os.path.join(logdir, 'plot_train'))
+        self.writer_valid = tf.summary.create_file_writer(os.path.join(logdir, 'plot_valid'))
+        self.writer_train_valid_loss = tf.summary.create_file_writer(os.path.join(logdir, 'plot_train_valid_loss'))
+
+        step = 0
+        total_time = 0
+        best_test_loss = np.finfo(dtype=np.float).max
+
+        if EARLY_STOP:
+            es = EarlyStop()
+
+        for epoch in range(NUM_EPOCHS):
+            self.train_loss.reset_states()
+            self.train_accuracy.reset_states()
+
+            t0 = datetime.datetime.now().timestamp()
+
+            proc_batch_cnt = 0
+            n_samples = 0
+
+            for data, label in self.train_dataset:
+                trn_ds = tf.data.Dataset.from_tensor_slices((data, label))
+                trn_ds = trn_ds.batch(BATCH_SIZE)
+                for mini_batch in trn_ds:
+                    if self.learningRateSchedule is not None:
+                        loss = self.train_step(mini_batch[0], mini_batch[1])
+
+                    if (step % 100) == 0:
+
+                        with self.writer_train.as_default():
+                            tf.summary.scalar('loss_trn', loss.numpy(), step=step)
+                            tf.summary.scalar('learning_rate', self.optimizer._decayed_lr('float32').numpy(), step=step)
+                            tf.summary.scalar('num_train_steps', step, step=step)
+                            tf.summary.scalar('num_epochs', epoch, step=step)
+
+                        self.reset_test_metrics()
+                        for data_tst, label_tst in self.test_dataset:
+                            tst_ds = tf.data.Dataset.from_tensor_slices((data_tst, label_tst))
+                            tst_ds = tst_ds.batch(BATCH_SIZE)
+                            for mini_batch_test in tst_ds:
+                                self.test_step(mini_batch_test[0], mini_batch_test[1])
+
+                        with self.writer_valid.as_default():
+                            tf.summary.scalar('loss_val', self.test_loss.result(), step=step)
+                            tf.summary.scalar('acc_val', self.test_accuracy.result(), step=step)
+
+                        with self.writer_train_valid_loss.as_default():
+                            tf.summary.scalar('loss_trn', loss.numpy(), step=step)
+                            tf.summary.scalar('loss_val', self.test_loss.result(), step=step)
+
+                        print('****** test loss, acc, lr: ', self.test_loss.result().numpy(), self.test_accuracy.result().numpy(),
+                              self.optimizer._decayed_lr('float32').numpy())
+
+                    step += 1
+                    print('train loss: ', loss.numpy())
+
+                proc_batch_cnt += 1
+                n_samples += data.shape[0]
+                print('proc_batch_cnt: ', proc_batch_cnt, n_samples)
+
+            t1 = datetime.datetime.now().timestamp()
+            print('End of Epoch: ', epoch+1, 'elapsed time: ', (t1-t0))
+            total_time += (t1-t0)
+
+            self.reset_test_metrics()
+            for data, label in self.test_dataset:
+                ds = tf.data.Dataset.from_tensor_slices((data, label))
+                ds = ds.batch(BATCH_SIZE)
+                for mini_batch in ds:
+                    self.test_step(mini_batch[0], mini_batch[1])
+
+            print('loss, acc: ', self.test_loss.result().numpy(), self.test_accuracy.result().numpy())
+            print('------------------------------------------------------')
+
+            tst_loss = self.test_loss.result().numpy()
+            if tst_loss < best_test_loss:
+                best_test_loss = tst_loss
+                ckpt_manager.save()
+
+            if EARLY_STOP and es.check_stop(tst_loss):
+                break
+
+        print('total time: ', total_time)
+        self.writer_train.close()
+        self.writer_valid.close()
+        self.writer_train_valid_loss.close()
+
+    def build_model(self):
+        self.build_srcnn()
+        self.model = tf.keras.Model(self.inputs, self.logits)
+
+    def restore(self, ckpt_dir):
+
+        ckpt = tf.train.Checkpoint(step=tf.Variable(1), model=self.model)
+        ckpt_manager = tf.train.CheckpointManager(ckpt, ckpt_dir, max_to_keep=3)
+        ckpt.restore(ckpt_manager.latest_checkpoint)
+
+        self.reset_test_metrics()
+
+        for data, label in self.test_dataset:
+            ds = tf.data.Dataset.from_tensor_slices((data, label))
+            ds = ds.batch(BATCH_SIZE)
+            for mini_batch_test in ds:
+                self.predict(mini_batch_test[0], mini_batch_test[1])
+
+        print('loss, acc: ', self.test_loss.result().numpy(), self.test_accuracy.result().numpy())
+
+        labels = np.concatenate(self.test_labels)
+        preds = np.concatenate(self.test_preds)
+        print(labels.shape, preds.shape)
+
+        labels = denormalize(labels, label_param, mean_std_dct)
+        preds = denormalize(preds, label_param, mean_std_dct)
+
+        return labels, preds
+
+    def do_evaluate(self, inputs, ckpt_dir):
+
+        ckpt = tf.train.Checkpoint(step=tf.Variable(1), model=self.model)
+        ckpt_manager = tf.train.CheckpointManager(ckpt, ckpt_dir, max_to_keep=3)
+        ckpt.restore(ckpt_manager.latest_checkpoint)
+
+        self.reset_test_metrics()
+
+        pred = self.model([inputs], training=False)
+        self.test_probs = pred
+        pred = pred.numpy()
+
+        return pred
+
+    def run(self, directory, ckpt_dir=None, num_data_samples=50000):
+        train_data_files = glob.glob(directory+'train*mres*.npy')
+        valid_data_files = glob.glob(directory+'valid*mres*.npy')
+        train_label_files = glob.glob(directory+'train*ires*.npy')
+        valid_label_files = glob.glob(directory+'valid*ires*.npy')
+        self.setup_pipeline(train_data_files, train_label_files, valid_data_files, valid_label_files, num_data_samples)
+
+        self.build_model()
+        self.build_training()
+        self.build_evaluation()
+        self.do_training(ckpt_dir=ckpt_dir)
+
+    def run_restore(self, directory, ckpt_dir):
+        self.num_data_samples = 1000
+
+        valid_data_files = glob.glob(directory + 'valid*mres*.npy')
+        valid_label_files = glob.glob(directory + 'valid*ires*.npy')
+        self.setup_test_pipeline(valid_data_files, valid_label_files)
+
+        self.build_model()
+        self.build_training()
+        self.build_evaluation()
+        return self.restore(ckpt_dir)
+
+    def run_evaluate(self, data, ckpt_dir):
+        # data = tf.convert_to_tensor(data, dtype=tf.float32)
+        self.num_data_samples = 80000
+        self.build_model()
+        self.build_training()
+        self.build_evaluation()
+        return self.do_evaluate(data, ckpt_dir)
+
+
+def run_restore_static(directory, ckpt_dir, out_file=None):
+    nn = SRCNN()
+    labels, preds = nn.run_restore(directory, ckpt_dir)
+    if out_file is not None:
+        np.save(out_file,
+                [np.squeeze(labels), preds.argmax(axis=3), preds[:, :, :, 0], preds[:, :, :, 1], preds[:, :, :, 2]])
+
+
+def run_evaluate_static(in_file, out_file, ckpt_dir):
+
+    h5f = h5py.File(in_file, 'r')
+
+    bt = get_grid_values_all(h5f, 'orig/temp_11_0um')
+    bt = np.where(np.isnan(bt), 0, bt)
+    bt = normalize(bt, 'temp_11_0um_nom', mean_std_dct)
+
+    refl = get_grid_values_all(h5f, 'super/refl_0_65um')
+    refl = np.where(np.isnan(refl), 0, refl)
+    refl = np.expand_dims(refl, axis=0)
+    refl_lo, refl_hi, refl_std, refl_avg = get_min_max_std(refl)
+    refl_lo = normalize(refl_lo, 'refl_0_65um_nom', mean_std_dct)
+    refl_hi = normalize(refl_hi, 'refl_0_65um_nom', mean_std_dct)
+    refl_avg = normalize(refl_avg, 'refl_0_65um_nom', mean_std_dct)
+    refl_lo = np.squeeze(refl_lo)
+    refl_hi = np.squeeze(refl_hi)
+    refl_avg = np.squeeze(refl_avg)
+
+    cp = get_grid_values_all(h5f, 'orig/'+label_param)
+    cp = np.where(np.isnan(cp), 0, cp)
+
+    data = np.stack([bt, refl_lo, refl_hi, refl_avg, cp], axis=2)
+    data = np.expand_dims(data, axis=0)
+
+    h5f.close()
+
+    nn = SRCNN()
+    probs = nn.run_evaluate(data, ckpt_dir)
+    cld_frac = probs.argmax(axis=3)
+
+    if out_file is not None:
+        np.save(out_file, (cld_frac[0, :, :], bt, refl_avg, cp))
+    else:
+        return cld_frac[0, :, :], bt, refl_avg, cp
+
+
+if __name__ == "__main__":
+    nn = SRCNN()
+    nn.run('matchup_filename')