Python datasets.fetch_mldata函数代码示例

 def __init__( self, use_mnist=False ):
   self.use_mnist = use_mnist
   if self.use_mnist:
     #self.digits = fetch_mldata('MNIST original')
     self.mnist_digits_train = fetch_mldata('MNIST original', subset='train')
     self.mnist_digits_test = fetch_mldata('MNIST original', subset='test')
   else:
     self.digits = load_digits()
   self.X = self.digits.data
   self.y = self.digits.target
   self.best_f1_score = 0
   self.best_score = 0
   """

def testScript():
	print "\n---> Started Logistic Regression - Iris dataset - Own function - k class...\n"
	attributes, outcomes = getDataFromFile("../Data/iriskc.data.shuffled")
	min_max_scaler = preprocessing.MinMaxScaler(feature_range=(-1,1))
	attributes, outcomes = min_max_scaler.fit_transform(np.array(attributes)), np.array(outcomes)
	#attributes, outcomes = np.array(attributes), np.array(outcomes)

	accrValues, presValues, recallValues, fMeasValues = crossValidate(attributes, outcomes, 10, learningRate=0.01, iterCountMax=750, threshold=0.005, ownFunction=True)
	for itr in range(10):
		print "Fold %d: \tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f" %(itr+1,accrValues[itr],presValues[itr],recallValues[itr],fMeasValues[itr])
	print "\nMean values:\tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f\n" % (np.mean(accrValues),np.mean(presValues),\
																												np.mean(recallValues),np.mean(fMeasValues))
															
																												
	print "---> Started Logistic Regression - Iris dataset - Inbuilt function - k class...\n"
	attributes, outcomes = getDataFromFile("../Data/iriskc.data.shuffled")
	min_max_scaler = preprocessing.MinMaxScaler(feature_range=(-1,1))
	attributes, outcomes = min_max_scaler.fit_transform(np.array(attributes)), np.array(outcomes)
	#attributes, outcomes = np.array(attributes), np.array(outcomes)

	accrValues, presValues, recallValues, fMeasValues = crossValidate(attributes, outcomes, 10, learningRate=0.01, iterCountMax=750, threshold=0.005, ownFunction=False)
	for itr in range(10):
		print "Fold %d: \tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f" %(itr+1,accrValues[itr],presValues[itr],recallValues[itr],fMeasValues[itr])
	print "\nMean values:\tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f\n" % (np.mean(accrValues),np.mean(presValues),\
																												np.mean(recallValues),np.mean(fMeasValues))


	print "---> Started Logistic Regression - Digits dataset - Own function - k class...\n"																									
	mnist = datasets.fetch_mldata('MNIST original')
	X, y = mnist.data / 255., mnist.target
	attributes = X[:20000]
	outcomes = y[:20000]
	#print list(set(outcomes))
	accrValues, presValues, recallValues, fMeasValues = crossValidate(attributes, outcomes, 10, learningRate=0.01, iterCountMax=100, threshold=0.005, ownFunction=False)
	for itr in range(10):
		print "Fold %d: \tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f" %(itr+1,accrValues[itr],presValues[itr],recallValues[itr],fMeasValues[itr])
	print "\nMean values:\tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f\n" % (np.mean(accrValues),np.mean(presValues),\
																												np.mean(recallValues),np.mean(fMeasValues))


	print "---> Started Logistic Regression - Digits dataset - Inbuilt function - k class...\n"																									
	mnist = datasets.fetch_mldata('MNIST original')
	X, y = mnist.data / 255., mnist.target
	attributes = X[:20000]
	outcomes = y[:20000]
	#print list(set(outcomes))
	accrValues, presValues, recallValues, fMeasValues = crossValidate(attributes, outcomes, 10, learningRate=0.01, iterCountMax=100, threshold=0.005, ownFunction=False)
	for itr in range(10):
		print "Fold %d: \tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f" %(itr+1,accrValues[itr],presValues[itr],recallValues[itr],fMeasValues[itr])
	print "\nMean values:\tAccuracy: %f\tPrecision: %f\tRecall: %f\tF-Measure: %f\n" % (np.mean(accrValues),np.mean(presValues),\
																												np.mean(recallValues),np.mean(fMeasValues))

def main():
    files = [
        join(SCRIPT_DIR, "train_x.npy"),
        join(SCRIPT_DIR, "train_y.npy"),
        join(SCRIPT_DIR, "validate_x.npy"),
        join(SCRIPT_DIR, "validate_y.npy"),
        join(SCRIPT_DIR, "test_x.npy"),
        join(SCRIPT_DIR, "test_y.npy")
    ]
    if all([exists(fname) and stat(fname).st_size > 100 for fname in files]):
        print("Already downloaded. Skipping")
    else:
        mnist = fetch_mldata('MNIST original')
        np.random.seed(1234)

        data = mnist.data
        target = mnist.target
        indices = np.arange(len(data))
        np.random.shuffle(indices)

        data = data[indices]
        target = target[indices]

        train_x, train_y = (data[:-10000].astype(np.float32) / 255.0).astype(np.float32), target[:-10000].astype(np.int32)
        test_x, test_y = (data[-10000:].astype(np.float32) / 255.0).astype(np.float32), target[-10000:].astype(np.int32)

        np.save(join(SCRIPT_DIR, "train_x.npy"), train_x[:int(0.9 * train_x.shape[0])])
        np.save(join(SCRIPT_DIR, "train_y.npy"), train_y[:int(0.9 * train_y.shape[0])])
        np.save(join(SCRIPT_DIR, "validate_x.npy"), train_x[int(0.9 * train_x.shape[0]):])
        np.save(join(SCRIPT_DIR, "validate_y.npy"), train_y[int(0.9 * train_y.shape[0]):])

        np.save(join(SCRIPT_DIR, "test_x.npy"), test_x)
        np.save(join(SCRIPT_DIR, "test_y.npy"), test_y)
        print("Done.")

def prepare_dataset():
    print('load MNIST dataset')
    mnist = fetch_mldata('MNIST original')
    mnist['data'] = mnist['data'].astype(np.float32)
    mnist['data'] /= 255
    mnist['target'] = mnist['target'].astype(np.int32)
    return mnist

def load(config, test=False):
    """Load MNIST dataset using scikit-learn.  Returns a dict with the
    following entries:

      - images:  n x 28 x 28 array
      - data:    n x 784  array
      - target:  n  array
    """
    dataset = fetch_mldata('mnist-original')
    X, y = dataset.data, dataset.target
    X = X.astype(np.float32) / 255.0

    if test:
        idx_start, idx_end = config['test_set']
    else:
        idx_start, idx_end = config['train_set']

    X, y = shuffle(X, y, random_state=42)
    X = X[idx_start:idx_end]
    y = y[idx_start:idx_end]

    return {
        'images': X.reshape(-1, 28, 28),
        'data': X,
        'target': y,
        }

def get_mnist(start=None, end=None, random=False, num=None):
    mnist = fetch_mldata('MNIST original', data_home='~/diss/mnist')
    if random is not None and num is not None:
        idx = np.random.choice(range(mnist.data.shape[0]), num)
    elif start is not None and end is not None:
        idx = range(start, end)
    return mnist.data[idx], mnist.target[idx]

def main(description, gpu, output):
	logging.basicConfig(level=logging.INFO)

	logging.info('fetch MNIST dataset')
	mnist = fetch_mldata(description)
	mnist.data = mnist.data.astype(numpy.float32)
	mnist.data /= 255
	mnist.target = mnist.target.astype(numpy.int32)

	data_train, data_test, target_train, target_test = train_test_split(mnist.data, mnist.target)

	data = data_train, data_test
	target = target_train, target_test

	start_time = time.time()

	if gpu >= 0:
		cuda.check_cuda_available()
		cuda.get_device(gpu).use()
		logging.info("Using gpu device {}".format(gpu))
	else:
		logging.info("Not using gpu device")

	mlp = MLP(data=data, target=target, gpu=gpu)
	mlp.train_and_test(n_epoch=1)

	end_time = time.time()

	logging.info("time = {} min".format((end_time - start_time) / 60.0))
	logging.info('saving trained mlp into {}'.format(output))
	with open(output, 'wb') as fp:
		pickle.dump(mlp, fp)

def test_classifier_chain_vs_independent_models():
    # Verify that an ensemble of classifier chains (each of length
    # N) can achieve a higher Jaccard similarity score than N independent
    # models
    yeast = fetch_mldata('yeast')
    X = yeast['data']
    Y = yeast['target'].transpose().toarray()
    X_train = X[:2000, :]
    X_test = X[2000:, :]
    Y_train = Y[:2000, :]
    Y_test = Y[2000:, :]

    ovr = OneVsRestClassifier(LogisticRegression())
    ovr.fit(X_train, Y_train)
    Y_pred_ovr = ovr.predict(X_test)

    chain = ClassifierChain(LogisticRegression(),
                            order=np.array([0, 2, 4, 6, 8, 10,
                                            12, 1, 3, 5, 7, 9,
                                            11, 13]))
    chain.fit(X_train, Y_train)
    Y_pred_chain = chain.predict(X_test)

    assert_greater(jaccard_similarity_score(Y_test, Y_pred_chain),
                   jaccard_similarity_score(Y_test, Y_pred_ovr))

def load_script(script_vars):
  def define(var_name, fun, overwrite=False):
    if script_vars.has_key(var_name) and not overwrite:
      print('%s is already defined' % var_name)
      return script_vars[var_name]
    else:
      print('computing variables %s' % var_name)
      value = fun()
      script_vars[var_name] = value
      globals()[var_name] = value      
      return value
  
    print(globals().keys())
    custom_data_home="/home/stefan2/mnistdata"

  custom_data_home="/home/stefan2/mnistdata"

  define('mnist', lambda: fetch_mldata('MNIST original', data_home=custom_data_home))
  data = mnist.data.astype(float) #[0:100,:] #convert to float
  labels = mnist.target #[0:100]
  n,m = data.shape
  print("num data points %s" % n)
  #run the method after successive othogonalization
  for j in range(0, 50):
    print("iteration: " + str(j))
    res = find_dominant_directions(data)
    plot_vector_png("pattern_" + str(j), res)
    for i in range(0, n):
      v = data[i,:]
      proj = np.reshape(v, (1, m)).dot(np.reshape(res, (m,1)))[0,0]
      data[i,:] = v - proj*res

def test_configs():    
    from sklearn import datasets
    from datetime import datetime
    import sys
    import os    
    import logging
    log = logging.getLogger()
    handler = logging.StreamHandler(sys.stdout)
    fmt = logging.Formatter('%(asctime)s %(levelname)s: %(message)s','%Y-%m-%d %H:%M:%S')
    handler.setFormatter(fmt)
    log.addHandler(handler)
    log.setLevel(logging.DEBUG)

    custom_data_home = os.getcwd() + '/sk_data'
    digits = datasets.fetch_mldata('MNIST original', data_home=custom_data_home)
    X = np.asarray(digits.data, 'float32')
    X = X
    # images = [imresize(im.reshape(28, 28), (32, 32)) for im in X]
    # X = np.vstack([im.flatten() for im in images])
    X[X < 128] = 0
    X[X >= 128] = 1
    X /= 256.
    models = []
    for w_sigma in [.1, .5, 1, 2, 5]:
        for sparsity in [.001, .01, .05, .1, .5]:
            log.info('Building RBM_dl:\n  w_sigma=%s\n  sparsity=%s' %(w_sigma,sparsity,))
            model = ConvRBM((28, 28), 40, w_size=11, n_iter=3, verbose=True, w_sigma=w_sigma, sparsity=sparsity)
            model.fit(X)
            models.append({
                'model' : model,
                'w_sigma' : w_sigma,
                'sparsity' : sparsity,
            })
    log.info('Done')
    return models

def run(data_path):
    print "Reading the dataset:", data_path

    ## http://continuum.io/blog/wiserf-use-cases-and-benchmarks

    mnist = fetch_mldata('MNIST original')

    # Define training and testing sets
    inds = arange(len(mnist.data))
    test_i = random.sample(xrange(len(inds)), int(0.1 * len(inds)))
    train_i = numpy.delete(inds, test_i)

    X_train = mnist.data[train_i].astype(numpy.double)
    y_train = mnist.target[train_i].astype(numpy.double)

    X_test = mnist.data[test_i].astype(numpy.double)
    y_test = mnist.target[test_i].astype(numpy.double)

    # Trunk the data
    X_digits, y_digits = shuffle(X_train, y_train)

    X_digits_train = X_digits[:1000]
    y_digits_train = y_digits[:1000]
    X_digits_valid = X_digits[1000:2000]
    y_digits_valid = y_digits[1000:2000]
    X_digits_test = X_digits[2000:3000]
    y_digits_test = y_digits[2000:3000]

    knn_digits = KNeighborsClassifier(n_neighbors=10)
    knn_digits.fit(X_digits_train, y_digits_train)
    print "KNN validation accuracy on MNIST digits: ",
    print knn_digits.score(X_digits_valid, y_digits_valid)

def main():
    print '... get mnist data'
    mnist = fetch_mldata('MNIST original', data_home='.')

    fig, axes = plt.subplots(5, 3, figsize=(6, 8))

    data = mnist.data[[0, 7000, 14000, 21000, 28000]]

    print '... start training'
    for i, (axrow, img) in enumerate(zip(axes, data)):
        img = img.reshape(28, 28)
        img = (img >= 128).astype(int)

        corrupted = get_corrupted_input(img, 0.05)
        mrf = MRF(corrupted)

        if i == 0:
            axes[i][0].set_title('元画像')
            axes[i][1].set_title('ノイズあり')
            axes[i][2].set_title('ノイズ除去')
        axes[i][0].imshow(img, cmap=cm.Greys_r)
        axes[i][1].imshow(corrupted, cmap=cm.Greys_r)
        axes[i][2].imshow(mrf.denoised, cmap=cm.Greys_r)
        for ax in axrow:
            ax.xaxis.set_visible(False)
            ax.yaxis.set_visible(False)
    plt.show()

def load(train_n, test_n):
    mnist = fetch_mldata('MNIST original', data_home='.')
    mnist.data = mnist.data.astype(np.float32) / 256.0
    mnist.target = mnist.target.astype(np.int32)
    N = len(mnist.data)

    order = np.random.permutation(N)

    train = {i: [] for i in range(10)}
    test = {i: [] for i in range(10)}

    train_m = math.ceil(train_n / 10)
    train_sum = 0

    test_m = math.ceil(test_n / 10)
    test_sum = 0

    for i in range(N):
        x = mnist.data[order[i]]
        y = mnist.target[order[i]]

        if train_sum < train_n and len(train[y]) < train_m:
            train[y].append(x)
            train_sum += 1

        if test_sum < test_n and len(test[y]) < test_m:
            test[y].append(x)
            test_sum += 1

    return train, test

 def download__by_category():
     # mnist = fetch_mldata('MNIST original')
     mnist = fetch_mldata('MNIST original')
     # mnist.data = random.sample(mnist.data, 1000)
     # mnist.target = random.sample(mnist.target, 1000)
     # mnist.data (70000, 784), mnist.target (70000, 1)
     trainX, trainY = mnist.data[:-10000], mnist.target[:-10000]
     testX, testY = mnist.data[-10000:], mnist.target[-10000:]
     if not exists('train'):
         os.makedirs('train')
     x = {i:[] for i in range(10)}
     for i in range(len(trainY)):
         tmp = x[trainY[i]]
         tmp.append(trainX[i])
         x[trainY[i]] = tmp
     for i in range(10):
         cPickle.dump(x[i], open(join('train', '{}.pkl'.format(i)), 'w+'))
     if not exists('test'):
         os.makedirs('test')
     x = {i:[] for i in range(10)}
     for i in range(len(testY)):
         tmp = x[testY[i]]
         tmp.append(testX[i])
         x[testY[i]] = tmp
     for i in range(10):
         cPickle.dump(x[i], open(join('test', '{}.pkl'.format(i)), 'w+'))

def make_data(N):
    print("fetch MNIST dataset")

    mnist = fetch_mldata('MNIST original',data_home='.')

    mnist.data = mnist.data.astype(np.float32)
    mnist.data /= 255

    mnist.taret = mnist.target.astype(np.int32)

    # make y label
    mnist_target = np.zeros((mnist.target.shape[0],10))

    for index, num in enumerate(mnist.target):
        mnist_target[index][num] = 1.

    # print(mnist_target)
    # mazemaze
    index = random.sample(range(mnist.target.shape[0]), (mnist.target.shape[0]))
    tmp_target = [mnist_target[i] for i in index]
    tmp_data = [mnist.data[i] for i in index]
    # print("N : ", len(tmp_target))
    # print("tmp_target : ", tmp_target)

    x_train, x_test = np.split(tmp_data, [N])
    y_train, y_test = np.split(tmp_target, [N])

    return [x_train, x_test, y_train, y_test]