Unlock 30% off on Manual Testing Annual Plans this Holiday Season.

Copied to Clipboard!

How to use start_step method in Lemoncheesecake

Best Python code snippet using lemoncheesecake

pruned_model.py

Source:pruned_model.py

1import tensorflow as tf2from tensorflow import keras3from tensorflow.python.keras.engine.base_layer import Layer4from tensorflow.keras.preprocessing.image import ImageDataGenerator5import numpy as np6import os7import cv28from tensorflow.python.framework import ops9from tensorflow.python.ops import math_ops10import tensorflow_model_optimization as tfmot11from tensorflow.keras.callbacks import TensorBoard12from tensorflow_model_optimization.python.core.sparsity.keras import pruning_wrapper13from tensorflow_model_optimization.python.core.quantization.keras import quantize_annotate14#tf.config.list_physical_devices('GPU')15# mish Activation16class Mish(Layer):17    '''18    Mish Activation Function.19    Examples:20        >>> X_input = Input(input_shape)21        >>> X = Mish()(X_input)22    '''23    def __init__(self, **kwargs):24        super(Mish, self).__init__(**kwargs)25        self.supports_masking = True26    def call(self, inputs):27        return inputs * keras.backend.tanh(keras.backend.log(keras.backend.exp(inputs)+1))28    def get_config(self):29        base_config = super(Mish, self).get_config()30        return {**base_config}31    def compute_output_shape(self, input_shape):32        return input_shape33class Sigmoid(Layer):34    def __init__(self, **kwargs):35        super(Sigmoid, self).__init__(**kwargs)36        self.supports_masking = True37    def call(self, inputs):38        return keras.backend.sigmoid(inputs)39    def get_config(self):40        base_config = super(Sigmoid, self).get_config()41        return {**base_config}42    def compute_output_shape(self, input_shape):43        return input_shape44# customize DNN modules45def backbone(input, start_step, end_step, prune, decay):46    conv = keras.layers.Conv2D(16, (3, 3), strides=(2, 2), padding='same')(input)47    bn = keras.layers.BatchNormalization(axis=-1, momentum=0.9)(conv)48    act = Mish()(bn)49    res = prune(keras.layers.Conv2D(16, (1, 1), strides=(2, 2), padding='same'), decay(0.02, 0.08, start_step, end_step))(act)50    bn_res = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.05, 0.15, start_step, end_step))(res)51    act_res = Mish()(bn_res)52    dws = prune(keras.layers.SeparableConv2D(16, (3, 3), padding='same'), decay(0.02, 0.08, start_step, end_step))(act)53    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.05, 0.15, start_step, end_step))(dws)54    act = Mish()(bn)55    dws = prune(keras.layers.SeparableConv2D(16, (3, 3), padding='same'), decay(0.05, 0.15, start_step, end_step))(act)56    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.1, 0.2, start_step, end_step))(dws)57    act = Mish()(bn)58    mp = keras.layers.MaxPooling2D((3, 3), strides=2, padding='same')(act)59    mp = keras.layers.add([mp, act_res])60    mp = Mish()(mp)61    conv = prune(keras.layers.Conv2D(32, (1, 1), strides=1, padding='same'), decay(0.15, 0.3, start_step, end_step))(mp)62    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.2, 0.4, start_step, end_step))(conv)63    act = Mish()(bn)64    conv = prune(keras.layers.Conv2D(64, (3, 3), strides=1, padding='same'), decay(0.15, 0.3, start_step, end_step))(bn)65    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.2, 0.4, start_step, end_step))(conv)66    act = Mish()(bn)67    conv_1 = prune(keras.layers.Conv2D(32, (1, 1), strides=1, padding='same'), decay(0.15, 0.3, start_step, end_step))(mp)68    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.2, 0.4, start_step, end_step))(conv_1)69    act_1 = Mish()(bn_1)70    conv_1 = prune(keras.layers.Conv2D(32, (5, 1), strides=1, padding='same'), decay(0.15, 0.3, start_step, end_step))(act_1)71    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.2, 0.4, start_step, end_step))(conv_1)72    act_1 = Mish()(bn_1)73    conv_1 = prune(keras.layers.Conv2D(32, (1, 5), strides=1, padding='same'), decay(0.15, 0.3, start_step, end_step))(act_1)74    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.2, 0.4, start_step, end_step))(conv_1)75    act_1 = Mish()(bn_1)76    conv_1 = prune(keras.layers.Conv2D(64, (3, 3), strides=1, padding='same'), decay(0.15, 0.3, start_step, end_step))(act_1)77    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.2, 0.4, start_step, end_step))(conv_1)78    act_1 = Mish()(bn_1)79    concat = tf.keras.layers.concatenate(inputs=[act, act_1], axis=3)80    return concat81def neck_module1(ip, start_step, end_step, prune, decay):82    input = Mish()(ip)83    conv = prune(keras.layers.Conv2D(32, (1, 1), strides=1, padding='same'), decay(0.2, 0.4, start_step, end_step))(input)84    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.4, 0.6, start_step, end_step))(conv)85    act = Mish()(bn)86    conv = prune(keras.layers.Conv2D(64, (1, 3), strides=1, padding='same'), decay(0.2, 0.4, start_step, end_step))(act)87    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.4, 0.6, start_step, end_step))(conv)88    act = Mish()(bn)89    conv = prune(keras.layers.Conv2D(32, (3, 1), strides=1, padding='same'), decay(0.2, 0.4, start_step, end_step))(act)90    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.4, 0.6, start_step, end_step))(conv)91    act = Mish()(bn)92    conv_1 = prune(keras.layers.Conv2D(32, (1, 1), strides=1, padding='same'), decay(0.3, 0.5, start_step, end_step))(input)93    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.5, 0.7, start_step, end_step))(conv_1)94    act_1 = Mish()(bn_1)95    conv_1 = prune(keras.layers.Conv2D(64, (1, 3), strides=1, padding='same'), decay(0.3, 0.5, start_step, end_step))(act_1)96    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.5, 0.7, start_step, end_step))(conv_1)97    act_1 = Mish()(bn_1)98    conv_1 = prune(keras.layers.Conv2D(64, (3, 1), strides=1, padding='same'), decay(0.3, 0.5, start_step, end_step))(act_1)99    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.5, 0.7, start_step, end_step))(conv_1)100    act_1 = Mish()(bn_1)101    conv_1 = prune(keras.layers.Conv2D(64, (1, 3), strides=1, padding='same'), decay(0.3, 0.5, start_step, end_step))(act_1)102    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.5, 0.7, start_step, end_step))(conv_1)103    act_1 = Mish()(bn_1)104    conv_1 = prune(keras.layers.Conv2D(32, (3, 1), strides=1, padding='same'), decay(0.3, 0.5, start_step, end_step))(act_1)105    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.5, 0.7, start_step, end_step))(conv_1)106    act_1 = Mish()(bn_1)107    mp = keras.layers.MaxPooling2D((2, 2), strides=1, padding='same')(input)108    conv_2 = prune(keras.layers.Conv2D(32, (1, 1), strides=1, padding='same'), decay(0.3, 0.5, start_step, end_step))(mp)109    bn_2 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.5, 0.7, start_step, end_step))(conv_2)110    act_2 = Mish()(bn_2)111    conv_3 = prune(keras.layers.Conv2D(32, (1, 1), strides=1, padding='same'), decay(0.3, 0.5, start_step, end_step))(input)112    bn_3 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.5, 0.7, start_step, end_step))(conv_3)113    act_3 = Mish()(bn_3)114    concat = keras.layers.concatenate(inputs=[act, act_1, act_2, act_3], axis=3)115    sum = keras.layers.add([concat, input])116    return sum117def neck_module2(ip, start_step, end_step, prune, decay):118    input = Mish()(ip)119    dws = prune(keras.layers.SeparableConv2D(64, (3, 3), padding='same'), decay(0.4, 0.7, start_step, end_step))(input)120    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.6, 0.8, start_step, end_step))(dws)121    act = Mish()(bn)122    dws = prune(keras.layers.SeparableConv2D(128, (3, 3), padding='same'), decay(0.4, 0.7, start_step, end_step))(act)123    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.6, 0.8, start_step, end_step))(dws)124    act = Mish()(bn)125    dws = prune(keras.layers.SeparableConv2D(192, (3, 3), padding='same'), decay(0.4, 0.7, start_step, end_step))(act)126    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.6, 0.8, start_step, end_step))(dws)127    act = Mish()(bn)128    dws_c = prune(keras.layers.SeparableConv2D(64, (3, 3), padding='same'), decay(0.4, 0.7, start_step, end_step))(input)129    bn_c = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.6, 0.8, start_step, end_step))(dws_c)130    act_c = Mish()(bn_c)131    concat = keras.layers.concatenate(inputs=[input, act_c], axis=3)132    sum = keras.layers.add([concat, act])133    act_sum = Mish()(sum)134    mp = keras.layers.MaxPooling2D((3, 3), strides=2, padding='same')(act_sum)135    conv = prune(keras.layers.Conv2D(128, (1, 1), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(mp)136    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv)137    act = Mish()(bn)138    conv = prune(keras.layers.Conv2D(128, (1, 3), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(act)139    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv)140    act = Mish()(bn)141    conv = prune(keras.layers.Conv2D(256, (3, 1), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(act)142    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv)143    act = Mish()(bn)144    conv_01 = prune(keras.layers.Conv2D(96, (1, 3), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(act)145    bn_01 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv_01)146    act_01 = Mish()(bn_01)147    conv_02 = prune(keras.layers.Conv2D(96, (3, 1), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(act)148    bn_02 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv_02)149    act_02 = Mish()(bn_02)150    conv_1 = prune(keras.layers.Conv2D(192, (1, 1), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(mp)151    bn_1 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv_1)152    act_1 = Mish()(bn_1)153    conv_11 = prune(keras.layers.Conv2D(96, (1, 3), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(act_1)154    bn_11 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv_11)155    act_11 = Mish()(bn_11)156    conv_12 = prune(keras.layers.Conv2D(96, (3, 1), strides=1, padding='same'), decay(0.6, 0.8, start_step, end_step))(act_1)157    bn_12 = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, start_step, end_step))(conv_12)158    act_12 = Mish()(bn_12)159    concat_1 = keras.layers.concatenate(inputs=[act_01, act_02], axis=3)160    concat_2 = keras.layers.concatenate(inputs=[act_11, act_12], axis=3)161    sum = keras.layers.add([mp, concat_1, concat_2])162    return sum163def head(ip, start_step, end_step, prune, decay):164    input = Mish()(ip)165    dws = prune(keras.layers.SeparableConv2D(256, (3, 3), padding='same'), decay(0.7, 0.9, 0, end_step))(input)166    bn_dws = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, 0, end_step))(dws)167    act_dws = Mish()(bn_dws)168    dws = prune(keras.layers.SeparableConv2D(384, (3, 3), padding='same'), decay(0.7, 0.9, 0, end_step))(act_dws)169    bn_dws = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, 0, end_step))(dws)170    act_dws = Mish()(bn_dws)171    conv = prune(keras.layers.Conv2D(384, (1, 1), strides=1, padding='same'), decay(0.7, 0.9, 0, end_step))(input)172    bn = prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, 0, end_step))(conv)173    act = Mish()(bn)174    sum = keras.layers.add([act, act_dws])175    act_sum = Mish()(sum)176    dws = prune(keras.layers.SeparableConv2D(512, (3, 3), padding='same'), decay(0.7, 0.9, 0, end_step))(act_sum)177    bn =  prune(keras.layers.BatchNormalization(axis=-1, momentum=0.9), decay(0.8, 0.9, 0, end_step))(dws)178    act = Mish()(bn)179    gap = keras.layers.GlobalAveragePooling2D()(act)180    return gap181def FADNet(input_shape, start_step, end_step, prune, decay):182    img_input = keras.layers.Input(shape=input_shape)183    bb = backbone(img_input, start_step, end_step, prune, decay)184    mp0 = keras.layers.MaxPooling2D((3, 3), strides=2, padding='same')(bb)185    n1 = neck_module1(mp0, start_step, end_step, prune, decay)186    mp1 = keras.layers.MaxPooling2D((3, 3), strides=2, padding='same')(n1)187    n2 = neck_module2(mp1, start_step, end_step, prune, decay)188    mp2 = keras.layers.MaxPooling2D((3, 3), strides=2, padding='same')(n2)189    h = head(mp2, start_step, end_step, prune, decay)190    d2 = prune(keras.layers.Dense(1), decay(0.7, 0.9, start_step, end_step))(h)191    d2 = Sigmoid()(d2)192    model = tf.keras.Model(img_input, d2)193    return model194# Customize early stopping callback195class MyThresholdCallback(tf.keras.callbacks.Callback):196    def __init__(self, tr_threshold, val_threshold):197        super(MyThresholdCallback, self).__init__()198        self.tr_threshold = tr_threshold199        self.val_threshold = val_threshold200    def on_epoch_end(self, epoch, logs=None):201        val_acc = logs["val_accuracy"]202        tr_acc = logs["accuracy"]203        if val_acc >= self.val_threshold or tr_acc >= self.tr_threshold:204            self.model.stop_training = True205# Customize learning rate schedule206class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):207    def __init__(self, initial_learning_rate, decay_steps, decay_rate, warmup_steps):208        super(CustomSchedule, self).__init__()209        self.initial_learning_rate = initial_learning_rate210        self.decay_steps = decay_steps211        self.decay_rate = decay_rate212        self.warmup_steps = warmup_steps213    def __call__(self, step):214        arg1 = self.initial_learning_rate * (self.decay_rate**((step-self.warmup_steps)/self.decay_steps))215        arg2 = self.initial_learning_rate * step / self.warmup_steps216        return tf.math.minimum(arg1, arg2)217    def get_config(self):218        return {219            "initial_learning_rate": self.initial_learning_rate,220            "decay_steps": self.decay_steps,221            "decay_rate": self.decay_rate,222            "warmup_steps": self.warmup_steps,223        }224def blur(img):225    return (cv2.blur(img,(5,5)))226def main():227    # Start k-fold cross vailidation228    k = 6229    val_acc = 0230    val_loss = 0231    model = None232    select = 1233    # Uncomment this to visualize an example of the learn rate schedule234    # temp_learning_rate_schedule = CustomSchedule(initial_learning_rate=0.001*0.95**50, decay_steps=584, decay_rate=0.95, warmup_steps=0)235    # plt.plot(temp_learning_rate_schedule(tf.range(40000, dtype=tf.float32)))236    # plt.ylabel("Learning Rate")237    # plt.xlabel("Train Step")238    # plt.show()239    fill_modes = ["constant", "reflect", "nearest"]240    rr = [10.0, 15.0, 10.0]241    sr = [10.0, 20.0, 10.0]242    cval = [np.random.random()*255, 0, 0]243    for i in range(3):244        # prefix = 'ks\\k' + str(i) + '\\'245        prefix = 'dataset/'246        # img_width, img_height = 100, 100247        train_data_dir = prefix + 'train/'248        validation_data_dir = prefix + 'test/'249        epochs = 30250        batch_size = 30251        if select == 0:252            select = 1253        elif select == 1:254            select = 2255        elif select == 2:256            select = 0257        print(fill_modes[select])258        # Generate Tensor for input images259        train_datagen = ImageDataGenerator(260            rescale=1. / 255,261            width_shift_range=0.2,262            height_shift_range=0.2,263            brightness_range=(0.7, 1.2),264            rotation_range=rr[select],265            shear_range=sr[select],266            zoom_range=(0.8, 1.3),267            channel_shift_range=60,268            fill_mode = fill_modes[select],269            cval=cval[select],270            horizontal_flip=True,271            vertical_flip=True)272        test_datagen = ImageDataGenerator(rescale=1. / 255,273                                          brightness_range=(0.9, 1.1),274                                          zoom_range=(1.0, 1.2),275                                          channel_shift_range=30,276                                          horizontal_flip=True)277        # Generate Iterator flow from directory278        train_generator = train_datagen.flow_from_directory(279            train_data_dir,280            # target_size=(img_width, img_height),281            target_size=(320, 320),282            batch_size=batch_size,283            class_mode='binary')284        validation_generator = test_datagen.flow_from_directory(285            validation_data_dir,286            # target_size=(img_width, img_height),287            target_size=(320, 320),288            batch_size=batch_size,289            class_mode='binary')290        # Get input image dimension291        unit_image = train_generator.next()[0]292        shape = (unit_image.shape[1], unit_image.shape[2], unit_image.shape[3])293        # pruning specifications294        poly_decay = tfmot.sparsity.keras.PolynomialDecay295        prune = tfmot.sparsity.keras.prune_low_magnitude296        start_step = 101297        end_step = int(584 * epochs / 2)298        quantize_annotate_layer = tfmot.quantization.keras.quantize_annotate_layer299        quantize_annotate_model = tfmot.quantization.keras.quantize_annotate_model300        quantize_scope = tfmot.quantization.keras.quantize_scope301        # Pruned DNN302        pruned_model = FADNet(shape, start_step, end_step, prune, poly_decay)303        # BP304        decay_exp = 0305        if i > 0:306            decay_exp = 60307        # if i == 5:308        #     th = 0.985309        lr_schedule = CustomSchedule(initial_learning_rate=0.0001*(0.98**decay_exp)/(i+1),310                                    decay_steps=584,311                                    decay_rate=0.95,312                                    warmup_steps=584*int((i+1)/2))313        opt = keras.optimizers.Adam(learning_rate=lr_schedule, clipnorm=5)314        bcel = keras.losses.BinaryCrossentropy(label_smoothing=0.1)315        # check for transfer learning316        if os.path.exists("quantized_model.h5"):317            pruned_model.load_weights("quantized_model.h5", by_name=True)318            print(pruned_model.layers[-2].get_weights()[0])319            print("Transfer learning continued")320        pruned_model.summary()321        pruned_model.compile(optimizer=opt, loss=bcel, metrics=['accuracy'])322        # callbacks323        my_callbacks = [324            tfmot.sparsity.keras.UpdatePruningStep(),325            # tfmot.sparsity.keras.PruningSummaries(log_dir='pruned_model'),326            keras.callbacks.EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True),327            MyThresholdCallback(tr_threshold=0.99, val_threshold=0.992+0.001*i),328            # TensorBoard(log_dir="prune"+str(i),329            #             histogram_freq=0,330            #             write_graph=True,331            #             write_images=False,332            #             update_freq="epoch",333            #             profile_batch=2),334            keras.callbacks.ModelCheckpoint(filepath='pruned_model.h5',335                                            save_weights_only=False,336                                            monitor='val_loss',337                                            mode='min',338                                            save_best_only=True)339        ]340        history = pruned_model.fit(train_generator, epochs=epochs, validation_data=validation_generator, callbacks=my_callbacks)341        # visualize compression342        # _, pruned_keras_file = tempfile.mkstemp('.h5')343        # tf.keras.models.save_model(model_for_export, pruned_keras_file, include_optimizer=False)344        # print("Size of gzipped pruned Keras model: %.2f bytes" % (get_gzipped_model_size(pruned_keras_file)))345        print('Finished part ' + str(i + 1))346        # strip pruning layers for final model347        model_for_export = tfmot.sparsity.keras.strip_pruning(pruned_model)348        model_for_export.save("pruned_model_stripped.h5")349        model_for_export.save_weights("pruned_model_weights.h5")350if __name__ == '__main__':...

trapping_water.py

Source:trapping_water.py

1class Solution:2    def trap(self, height):3        if len(height) < 3:4            return []5        my_trap = []6        start_step = 07        while start_step <= len(height) - 3:8            if height[start_step] > height[start_step + 1]:9                j = start_step + 110                trap_sequence = []11                trap_sequence.append(height[start_step])12                while j < len(height) -1 and height[j] < height[start_step]:13                    trap_sequence.append(height[j])14                    j += 115                if height[j] >= height[start_step]:16                    trap_sequence.append(height[j])17                else:18                    trap_sequence = []19                if len(trap_sequence) > 0:20                    my_trap.append(trap_sequence)21                start_step = j22            else:23                start_step +=124        return self.calculate(my_trap)25    def calculate(self, trap):26        trapped_water = 027        for trap_sequence in trap:28            sum  = 029            mark_heigh = trap_sequence[0]30            for j in trap_sequence:31                sum += max(mark_heigh - j, 0)32            trapped_water += sum33        return trapped_water34if __name__ == '__main__':35    s = Solution()...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.