Keras自定义拟合生成器y_true

时间:2019-03-14 17:03:54

标签: machine-learning keras generator

我正在使用data.py中定义的自定义帧生成器:

import csv
import numpy as np
import random
import glob
import os.path
import sys
import operator
import threading
from processor import process_image
from keras.utils import to_categorical

class threadsafe_iterator:
    def __init__(self, iterator):
        self.iterator = iterator
        self.lock = threading.Lock()

    def __iter__(self):
        return self

    def next(self):
        with self.lock:
            return next(self.iterator)

def threadsafe_generator(func):
    """Decorator"""
    def gen(*a, **kw):
        return threadsafe_iterator(func(*a, **kw))
    return gen

class DataSet():

    def __init__(self, seq_length=40, class_limit=None, image_shape=(80, 80, 3)):
        """Constructor.
        seq_length = (int) the number of frames to consider
        class_limit = (int) number of classes to limit the data to.
            None = no limit.
        """
        self.seq_length = seq_length
        self.class_limit = class_limit
        self.sequence_path = os.path.join('data', 'sequences')
        self.max_frames = 300  # max number of frames a video can have for us to use it

        # Get the data.
        self.data = self.get_data()

        # Get the classes.
        self.classes = self.get_classes()

        # Now do some minor data cleaning.
        self.data = self.clean_data()

        self.image_shape = image_shape

    @staticmethod
    def get_data():
        """Load our data from file."""
        with open(os.path.join('data', 'data_file.csv'), 'r') as fin:
            reader = csv.reader(fin)
            data = list(reader)

        return data

    def clean_data(self):
        """Limit samples to greater than the sequence length and fewer
        than N frames. Also limit it to classes we want to use."""
        data_clean = []
        for item in self.data:
            if int(item[3]) >= self.seq_length and int(item[3]) <= self.max_frames \
                    and item[1] in self.classes:
                data_clean.append(item)

        return data_clean

    def get_classes(self):
        """Extract the classes from our data. If we want to limit them,
        only return the classes we need."""
        classes = []
        for item in self.data:
            if item[1] not in classes:
                classes.append(item[1])

        # Sort them.
        classes = sorted(classes)

        # Return.
        if self.class_limit is not None:
            return classes[:self.class_limit]
        else:
            return classes

    def get_class_one_hot(self, class_str):
        """Given a class as a string, return its number in the classes
        list. This lets us encode and one-hot it for training."""
        # Encode it first.
        label_encoded = self.classes.index(class_str)

        # Now one-hot it.
        label_hot = to_categorical(label_encoded, len(self.classes))

        assert len(label_hot) == len(self.classes)

        return label_hot

    def split_train_test(self):
        """Split the data into train and test groups."""
        train = []
        test = []
        for item in self.data:

            if item[0] == 'training':
                train.append(item)
            else:
                test.append(item)
        return train, test

    def get_train_length(self):
        train, test = split_train_test()
        return len(train)

    def get_test_length(self):
        train, test = split_train_test()
        return len(test)

    def get_all_sequences_in_memory(self, train_test, data_type):
        """
        This is a mirror of our generator, but attempts to load everything into
        memory so we can train way faster.
        """
        # Get the right dataset.
        train, test = self.split_train_test()
        data = train if train_test == 'training' else test

        print("Loading %d samples into memory for %sing." % (len(data), train_test))

        X, y = [], []
        for row in data:

            if data_type == 'images':
                frames = self.get_frames_for_sample(row)
                frames = self.rescale_list(frames, self.seq_length)

                # Build the image sequence
                sequence = self.build_image_sequence(frames)

            else:
                sequence = self.get_extracted_sequence(data_type, row)

                if sequence is None:
                    print("Can't find sequence. Did you generate them?")
                    raise

            X.append(sequence)
            y.append(self.get_class_one_hot(row[1]))

        return np.array(X), np.array(y)

    @threadsafe_generator
    def frame_generator(self, batch_size, train_test, data_type):
        """Return a generator that we can use to train on. There are
        a couple different things we can return:

        data_type: 'features', 'images'
        """
        # Get the right dataset for the generator.
        train, test = self.split_train_test()
        data = train if train_test == 'training' else test

        print("Creating %s generator with %d samples." % (train_test, len(data)))

        while 1:
            X, y = [], []

            # Generate batch_size samples.
            for _ in range(batch_size):
                # Reset to be safe.
                sequence = None

                # Get a random sample.
                sample = random.choice(data)
                # Check to see if we've already saved this sequence.
                if data_type is "images":
                    # Get and resample frames.
                    frames = self.get_frames_for_sample(sample)
                    frames = self.rescale_list(frames, self.seq_length)

                    # Build the image sequence
                    sequence = self.build_image_sequence(frames)
                else:
                    # Get the sequence from disk.
                    sequence = self.get_extracted_sequence(data_type, sample)

                    if sequence is None:
                        raise ValueError("Can't find sequence. Did you generate them?")

                X.append(sequence)
                y.append(self.get_class_one_hot(sample[1]))

            yield np.array(X), np.array(y)

    def build_image_sequence(self, frames):
        """Given a set of frames (filenames), build our sequence."""

        return [process_image(x, self.image_shape) for x in frames]

    def get_extracted_sequence(self, data_type, sample):
        """Get the saved extracted features."""
        filename = sample[2]
        path = os.path.join(self.sequence_path, filename + '-' + str(self.seq_length) + \
            '-' + data_type + '.npy')
        if os.path.isfile(path):
            return np.load(path)
        else:
            return None

    def get_frames_by_filename(self, filename, data_type):
        """Given a filename for one of our samples, return the data
        the model needs to make predictions."""
        # First, find the sample row.
        sample = None
        for row in self.data:
            if row[2] == filename:
                sample = row
                break
        if sample is None:
            raise ValueError("Couldn't find sample: %s" % filename)

        if data_type == "images":
            # Get and resample frames.
            frames = self.get_frames_for_sample(sample)
            frames = self.rescale_list(frames, self.seq_length)
            # Build the image sequence
            sequence = self.build_image_sequence(frames)
        else:
            # Get the sequence from disk.
            sequence = self.get_extracted_sequence(data_type, sample)

            if sequence is None:
                raise ValueError("Can't find sequence. Did you generate them?")

        return sequence

    @staticmethod
    def get_frames_for_sample(sample):

        """Given a sample row from the data file, get all the corresponding frame
        filenames."""
        path = os.path.join('/home/john/medical_decrypted/3d_cnn/Data_Videos', sample[0], sample[1])
        filename = sample[2]
        images = sorted(glob.glob(os.path.join(path, filename + '*.jpg')))


        return images

    @staticmethod
    def get_filename_from_image(filename):
        parts = filename.split(os.path.sep)
        return parts[-1].replace('.jpg', '')

    @staticmethod
    def rescale_list(input_list, size):
        """Given a list and a size, return a rescaled/samples list. For example,
        if we want a list of size 5 and we have a list of size 25, return a new
        list of size five which is every 5th element of the origina list."""
        assert len(input_list) >= size

        # Get the number to skip between iterations.
        skip = len(input_list) // size

        # Build our new output.
        output = [input_list[i] for i in range(0, len(input_list), skip)]

        # Cut off the last one if needed.
        return output[:size]

    def print_class_from_prediction(self, predictions, nb_to_return=5):
        """Given a prediction, print the top classes."""
        # Get the prediction for each label.
        label_predictions = {}
        for i, label in enumerate(self.classes):
            label_predictions[label] = predictions[i]

        # Now sort them.
        sorted_lps = sorted(
            label_predictions.items(),
            key=operator.itemgetter(1),
            reverse=True
        )

        # And return the top N.
        for i, class_prediction in enumerate(sorted_lps):
            if i > nb_to_return - 1 or class_prediction[1] == 0.0:
                break
            print("%s: %.2f" % (class_prediction[0], class_prediction[1]))

在train.py中用于训练3D CNN:

"""
Train our RNN on extracted features or images.
"""
from models import ResearchModels
from data import DataSet
import time
import os.path
from keras.callbacks import TensorBoard, ModelCheckpoint, EarlyStopping, CSVLogger
from sklearn.metrics import classification_report 
import numpy as np

np.set_printoptions(threshold=np.nan)

def train(data_type, seq_length, model, saved_model=None,
          class_limit=None, image_shape=None,
          load_to_memory=False, batch_size=32, nb_epoch=100):
    # Helper: Save the model.
    checkpointer = ModelCheckpoint(
        filepath=os.path.join('data', 'checkpoints', model + '-' + data_type + \
            '.{epoch:03d}-{val_loss:.3f}.hdf5'),
        verbose=1,
        save_best_only=True)

    # Helper: TensorBoard
    tb = TensorBoard(log_dir=os.path.join('data', 'logs', model))

    # Helper: Stop when we stop learning.
    early_stopper = EarlyStopping(patience=5)

    # Helper: Save results.
    timestamp = time.time()
    csv_logger = CSVLogger(os.path.join('data', 'logs', model + '-' + 'training-' + \
        str(timestamp) + '.log'))

    # Get the data and process it.
    if image_shape is None:
        data = DataSet(
            seq_length=seq_length,
            class_limit=class_limit
        )
    else:
        data = DataSet(
            seq_length=seq_length,
            class_limit=class_limit,
            image_shape=image_shape
        )
        data.data = data.get_data()

    # Get samples per epoch.
    # Multiply by 0.7 to attempt to guess how much of data.data is the train set.

    steps_per_epoch = data.get_train_length
    validation_steps = data.get_test_length 
    print(steps_per_epoch)
    print(validation_steps)


    if load_to_memory:
        # Get data.
        X, y = data.get_all_sequences_in_memory('training', data_type)
        X_test, y_test = data.get_all_sequences_in_memory('test', data_type)
    else:
        # Get generators.
        generator = data.frame_generator(batch_size, 'training', data_type)
        val_generator = data.frame_generator(batch_size, 'test', data_type)
    # Get the model.
    rm = ResearchModels(len(data.classes), model, seq_length, saved_model)


    if load_to_memory:
        # Use standard fit.
        rm.model.fit(
            X,
            y,
            batch_size=batch_size,
            validation_data=(X_test, y_test),
            verbose=1,
            callbacks=[tb, early_stopper, csv_logger],
            epochs=nb_epoch)
    else:
        # Use fit generator.
        history = rm.model.fit_generator(
            generator=generator,
            steps_per_epoch=steps_per_epoch,
            epochs=1,
            verbose=1,
            callbacks=[tb, early_stopper, csv_logger, checkpointer],
            validation_data=val_generator,
            validation_steps=validation_steps,
            workers=2)

        print(history.history)
        print("[INFO] Get Predictions")
        predictions = rm.model.predict_generator(val_generator, 1, verbose=1)
        y_preds = np.argmax(predictions, axis=-1) 

        # y_true = val_generator.argmax(axis=-1)
        # cr = classification_report(y_true, y_preds)


def main():
    """These are the main training settings. Set each before running
    this file."""
    # model can be one of lstm, lrcn, mlp, conv_3d, c3d
    model = 'conv_3d'
    saved_model = None  # None or weights file
    class_limit = None  # int, can be 1-101 or None
    seq_length = 40
    load_to_memory = False  # pre-load the sequences into memory
    batch_size = 3
    nb_epoch = 1

    # Chose images or features and image shape based on network.
    if model in ['conv_3d', 'c3d', 'lrcn']:
        data_type = 'images'
        image_shape = (80, 80, 3)
    elif model in ['lstm', 'mlp']:
        data_type = 'features'
        image_shape = None
    else:
        raise ValueError("Invalid model. See train.py for options.")

    train(data_type, seq_length, model, saved_model=saved_model,
          class_limit=class_limit, image_shape=image_shape,
          load_to_memory=load_to_memory, batch_size=batch_size, nb_epoch=nb_epoch)

if __name__ == '__main__':
    main()

我的问题是如何获取与生成器做出的预测相对应的y_true值,以便可以调用sklearn分类报告功能?我还需要它来计算ROC曲线。

0 个答案:

没有答案