caffe示例实现之11逻辑回归及其深化

虽然Caffe是设计给深度网络的，但它也可以表示“浅层”网络模型用于分类，例如逻辑回归。下面在合成数据上做简单的逻辑回归，生成并保存为HDF5，来向Caffe输入向量。模型完成后，添加层来提高准确率。Caffe完成的主要内容就是：定义模型，实验，应用。

import numpy as npimport matplotlib.pyplot as plt%matplotlib inlineimport osos.chdir('..')import syssys.path.insert(0, './python')import caffeimport osimport h5pyimport shutilimport tempfileimport sklearnimport sklearn.datasetsimport sklearn.linear_modelimport pandas as pd

合成10,000个4向量的数据集，用于有2个有用特征和2个噪声特征的二值分类。

X, y = sklearn.datasets.make_classification(    n_samples=10000, n_features=4, n_redundant=0, n_informative=2,     n_clusters_per_class=2, hypercube=False, random_state=0)# Split into train and testX, Xt, y, yt = sklearn.cross_validation.train_test_split(X, y)# Visualize sample of the dataind = np.random.permutation(X.shape[0])[:1000]df = pd.DataFrame(X[ind])_ = pd.scatter_matrix(df, figsize=(9, 9), diagonal='kde', marker='o', s=40, alpha=.4, c=y[ind])

用于学习scikit-learn逻辑回归的方法是随机梯度下降（SGD），下面检测分类器的准确性：

%%timeit# Train and test the scikit-learn SGD logistic regression.clf = sklearn.linear_model.SGDClassifier(    loss='log', n_iter=1000, penalty='l2', alpha=1e-3, class_weight='auto')clf.fit(X, y)yt_pred = clf.predict(Xt)print('Accuracy: {:.3f}'.format(sklearn.metrics.accuracy_score(yt, yt_pred)))

Accuracy: 0.783
Accuracy: 0.783
Accuracy: 0.783
Accuracy: 0.783
1 loops, best of 3: 508 ms per loop

将这个数据集保存到HDF5中，方便Caffe载入：

# Write out the data to HDF5 files in a temp directory.# This file is assumed to be caffe_root/examples/hdf5_classification.ipynbdirname = os.path.abspath('./examples/hdf5_classification/data')if not os.path.exists(dirname):    os.makedirs(dirname)train_filename = os.path.join(dirname, 'train.h5')test_filename = os.path.join(dirname, 'test.h5')# HDF5DataLayer source should be a file containing a list of HDF5 filenames.# To show this off, we'll list the same data file twice.with h5py.File(train_filename, 'w') as f:    f['data'] = X    f['label'] = y.astype(np.float32)with open(os.path.join(dirname, 'train.txt'), 'w') as f:    f.write(train_filename + '\n')    f.write(train_filename + '\n')# HDF5 is pretty efficient, but can be further compressed.comp_kwargs = {'compression': 'gzip', 'compression_opts': 1}with h5py.File(test_filename, 'w') as f:    f.create_dataset('data', data=Xt, **comp_kwargs)    f.create_dataset('label', data=yt.astype(np.float32), **comp_kwargs)with open(os.path.join(dirname, 'test.txt'), 'w') as f:    f.write(test_filename + '\n')

通过Python网络规范来定义Caffe中的逻辑回归，如果不想手工编写protobuf模型的话这是一个好方法：

from caffe import layers as Lfrom caffe import params as Pdef logreg(hdf5, batch_size):    # logistic regression: data, matrix multiplication, and 2-class softmax loss    n = caffe.NetSpec()    n.data, n.label = L.HDF5Data(batch_size=batch_size, source=hdf5, ntop=2)    n.ip1 = L.InnerProduct(n.data, num_output=2, weight_filler=dict(type='xavier'))    n.accuracy = L.Accuracy(n.ip1, n.label)    n.loss = L.SoftmaxWithLoss(n.ip1, n.label)    return n.to_proto()with open('examples/hdf5_classification/logreg_auto_train.prototxt', 'w') as f:    f.write(str(logreg('examples/hdf5_classification/data/train.txt', 10)))with open('examples/hdf5_classification/logreg_auto_test.prototxt', 'w') as f:    f.write(str(logreg('examples/hdf5_classification/data/test.txt', 10)))

现在开始学习与评估Python中的Caffe逻辑回归：

%%timeitcaffe.set_mode_cpu()solver = caffe.get_solver('examples/hdf5_classification/solver.prototxt')solver.solve()accuracy = 0batch_size = solver.test_nets[0].blobs['data'].numtest_iters = int(len(Xt) / batch_size)for i in range(test_iters):    solver.test_nets[0].forward()    accuracy += solver.test_nets[0].blobs['accuracy'].dataaccuracy /= test_itersprint("Accuracy: {:.3f}".format(accuracy))

Accuracy: 0.782
Accuracy: 0.782
Accuracy: 0.782
Accuracy: 0.782
1 loops, best of 3: 287 ms per loop

用命令行来运行，看一下具体的模型文件和求解过程：

!./build/tools/caffe train -solver examples/hdf5_classification/solver.prototxt

I0318 00:58:32.322571 2013098752 caffe.cpp:117] Use CPU.I0318 00:58:32.643163 2013098752 caffe.cpp:121] Starting OptimizationI0318 00:58:32.643229 2013098752 solver.cpp:32] Initializing solver from parameters: train_net: "examples/hdf5_classification/logreg_auto_train.prototxt"test_net: "examples/hdf5_classification/logreg_auto_test.prototxt"test_iter: 250test_interval: 1000base_lr: 0.01display: 1000max_iter: 10000lr_policy: "step"gamma: 0.1momentum: 0.9weight_decay: 0.0005stepsize: 5000snapshot: 10000snapshot_prefix: "examples/hdf5_classification/data/train"solver_mode: CPUI0318 00:58:32.643333 2013098752 solver.cpp:61] Creating training net from train_net file: examples/hdf5_classification/logreg_auto_train.prototxtI0318 00:58:32.643465 2013098752 net.cpp:42] Initializing net from parameters: state {  phase: TRAIN}layer {  name: "data"  type: "HDF5Data"  top: "data"  top: "label"  hdf5_data_param {    source: "examples/hdf5_classification/data/train.txt"    batch_size: 10  }}layer {  name: "ip1"  type: "InnerProduct"  bottom: "data"  top: "ip1"  inner_product_param {    num_output: 2    weight_filler {      type: "xavier"    }  }}layer {  name: "accuracy"  type: "Accuracy"  bottom: "ip1"  bottom: "label"  top: "accuracy"}layer {  name: "loss"  type: "SoftmaxWithLoss"  bottom: "ip1"  bottom: "label"  top: "loss"}I0318 00:58:32.644197 2013098752 layer_factory.hpp:74] Creating layer dataI0318 00:58:32.644219 2013098752 net.cpp:84] Creating Layer dataI0318 00:58:32.644230 2013098752 net.cpp:338] data -> dataI0318 00:58:32.644256 2013098752 net.cpp:338] data -> labelI0318 00:58:32.644269 2013098752 net.cpp:113] Setting up dataI0318 00:58:32.644278 2013098752 hdf5_data_layer.cpp:66] Loading list of HDF5 filenames from: examples/hdf5_classification/data/train.txtI0318 00:58:32.644327 2013098752 hdf5_data_layer.cpp:80] Number of HDF5 files: 2I0318 00:58:32.646458 2013098752 net.cpp:120] Top shape: 10 4 (40)I0318 00:58:32.646502 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:32.646518 2013098752 layer_factory.hpp:74] Creating layer label_data_1_splitI0318 00:58:32.646538 2013098752 net.cpp:84] Creating Layer label_data_1_splitI0318 00:58:32.646546 2013098752 net.cpp:380] label_data_1_split <- labelI0318 00:58:32.646556 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_0I0318 00:58:32.646569 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_1I0318 00:58:32.646579 2013098752 net.cpp:113] Setting up label_data_1_splitI0318 00:58:32.646586 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:32.646595 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:32.646601 2013098752 layer_factory.hpp:74] Creating layer ip1I0318 00:58:32.646615 2013098752 net.cpp:84] Creating Layer ip1I0318 00:58:32.646622 2013098752 net.cpp:380] ip1 <- dataI0318 00:58:32.646664 2013098752 net.cpp:338] ip1 -> ip1I0318 00:58:32.646689 2013098752 net.cpp:113] Setting up ip1I0318 00:58:32.652330 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:32.652371 2013098752 layer_factory.hpp:74] Creating layer ip1_ip1_0_splitI0318 00:58:32.652393 2013098752 net.cpp:84] Creating Layer ip1_ip1_0_splitI0318 00:58:32.652407 2013098752 net.cpp:380] ip1_ip1_0_split <- ip1I0318 00:58:32.652421 2013098752 net.cpp:338] ip1_ip1_0_split -> ip1_ip1_0_split_0I0318 00:58:32.652467 2013098752 net.cpp:338] ip1_ip1_0_split -> ip1_ip1_0_split_1I0318 00:58:32.652480 2013098752 net.cpp:113] Setting up ip1_ip1_0_splitI0318 00:58:32.652489 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:32.652498 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:32.652505 2013098752 layer_factory.hpp:74] Creating layer accuracyI0318 00:58:32.652521 2013098752 net.cpp:84] Creating Layer accuracyI0318 00:58:32.652534 2013098752 net.cpp:380] accuracy <- ip1_ip1_0_split_0I0318 00:58:32.652545 2013098752 net.cpp:380] accuracy <- label_data_1_split_0I0318 00:58:32.652562 2013098752 net.cpp:338] accuracy -> accuracyI0318 00:58:32.652577 2013098752 net.cpp:113] Setting up accuracyI0318 00:58:32.652590 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:32.652642 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:32.652655 2013098752 net.cpp:84] Creating Layer lossI0318 00:58:32.652663 2013098752 net.cpp:380] loss <- ip1_ip1_0_split_1I0318 00:58:32.652672 2013098752 net.cpp:380] loss <- label_data_1_split_1I0318 00:58:32.652679 2013098752 net.cpp:338] loss -> lossI0318 00:58:32.652689 2013098752 net.cpp:113] Setting up lossI0318 00:58:32.652701 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:32.652716 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:32.652724 2013098752 net.cpp:122]     with loss weight 1I0318 00:58:32.652740 2013098752 net.cpp:167] loss needs backward computation.I0318 00:58:32.652746 2013098752 net.cpp:169] accuracy does not need backward computation.I0318 00:58:32.652753 2013098752 net.cpp:167] ip1_ip1_0_split needs backward computation.I0318 00:58:32.652760 2013098752 net.cpp:167] ip1 needs backward computation.I0318 00:58:32.652786 2013098752 net.cpp:169] label_data_1_split does not need backward computation.I0318 00:58:32.652801 2013098752 net.cpp:169] data does not need backward computation.I0318 00:58:32.652808 2013098752 net.cpp:205] This network produces output accuracyI0318 00:58:32.652815 2013098752 net.cpp:205] This network produces output lossI0318 00:58:32.652825 2013098752 net.cpp:447] Collecting Learning Rate and Weight Decay.I0318 00:58:32.652833 2013098752 net.cpp:217] Network initialization done.I0318 00:58:32.652839 2013098752 net.cpp:218] Memory required for data: 528I0318 00:58:32.652964 2013098752 solver.cpp:154] Creating test net (#0) specified by test_net file: examples/hdf5_classification/logreg_auto_test.prototxtI0318 00:58:32.652986 2013098752 net.cpp:42] Initializing net from parameters: state {  phase: TEST}layer {  name: "data"  type: "HDF5Data"  top: "data"  top: "label"  hdf5_data_param {    source: "examples/hdf5_classification/data/test.txt"    batch_size: 10  }}layer {  name: "ip1"  type: "InnerProduct"  bottom: "data"  top: "ip1"  inner_product_param {    num_output: 2    weight_filler {      type: "xavier"    }  }}layer {  name: "accuracy"  type: "Accuracy"  bottom: "ip1"  bottom: "label"  top: "accuracy"}layer {  name: "loss"  type: "SoftmaxWithLoss"  bottom: "ip1"  bottom: "label"  top: "loss"}I0318 00:58:32.653069 2013098752 layer_factory.hpp:74] Creating layer dataI0318 00:58:32.653080 2013098752 net.cpp:84] Creating Layer dataI0318 00:58:32.653090 2013098752 net.cpp:338] data -> dataI0318 00:58:32.653128 2013098752 net.cpp:338] data -> labelI0318 00:58:32.653146 2013098752 net.cpp:113] Setting up dataI0318 00:58:32.653154 2013098752 hdf5_data_layer.cpp:66] Loading list of HDF5 filenames from: examples/hdf5_classification/data/test.txtI0318 00:58:32.653192 2013098752 hdf5_data_layer.cpp:80] Number of HDF5 files: 1I0318 00:58:32.654850 2013098752 net.cpp:120] Top shape: 10 4 (40)I0318 00:58:32.654897 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:32.654914 2013098752 layer_factory.hpp:74] Creating layer label_data_1_splitI0318 00:58:32.654933 2013098752 net.cpp:84] Creating Layer label_data_1_splitI0318 00:58:32.654943 2013098752 net.cpp:380] label_data_1_split <- labelI0318 00:58:32.654953 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_0I0318 00:58:32.654966 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_1I0318 00:58:32.654976 2013098752 net.cpp:113] Setting up label_data_1_splitI0318 00:58:32.654985 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:32.654992 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:32.655000 2013098752 layer_factory.hpp:74] Creating layer ip1I0318 00:58:32.655010 2013098752 net.cpp:84] Creating Layer ip1I0318 00:58:32.655017 2013098752 net.cpp:380] ip1 <- dataI0318 00:58:32.655030 2013098752 net.cpp:338] ip1 -> ip1I0318 00:58:32.655041 2013098752 net.cpp:113] Setting up ip1I0318 00:58:32.655061 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:32.655072 2013098752 layer_factory.hpp:74] Creating layer ip1_ip1_0_splitI0318 00:58:32.655148 2013098752 net.cpp:84] Creating Layer ip1_ip1_0_splitI0318 00:58:32.655159 2013098752 net.cpp:380] ip1_ip1_0_split <- ip1I0318 00:58:32.655170 2013098752 net.cpp:338] ip1_ip1_0_split -> ip1_ip1_0_split_0I0318 00:58:32.655180 2013098752 net.cpp:338] ip1_ip1_0_split -> ip1_ip1_0_split_1I0318 00:58:32.655190 2013098752 net.cpp:113] Setting up ip1_ip1_0_splitI0318 00:58:32.655199 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:32.655206 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:32.655213 2013098752 layer_factory.hpp:74] Creating layer accuracyI0318 00:58:32.655223 2013098752 net.cpp:84] Creating Layer accuracyI0318 00:58:32.655230 2013098752 net.cpp:380] accuracy <- ip1_ip1_0_split_0I0318 00:58:32.655237 2013098752 net.cpp:380] accuracy <- label_data_1_split_0I0318 00:58:32.655251 2013098752 net.cpp:338] accuracy -> accuracyI0318 00:58:32.655259 2013098752 net.cpp:113] Setting up accuracyI0318 00:58:32.655267 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:32.655340 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:32.655354 2013098752 net.cpp:84] Creating Layer lossI0318 00:58:32.655361 2013098752 net.cpp:380] loss <- ip1_ip1_0_split_1I0318 00:58:32.655369 2013098752 net.cpp:380] loss <- label_data_1_split_1I0318 00:58:32.655378 2013098752 net.cpp:338] loss -> lossI0318 00:58:32.655388 2013098752 net.cpp:113] Setting up lossI0318 00:58:32.655397 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:32.655414 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:32.655422 2013098752 net.cpp:122]     with loss weight 1I0318 00:58:32.655438 2013098752 net.cpp:167] loss needs backward computation.I0318 00:58:32.655446 2013098752 net.cpp:169] accuracy does not need backward computation.I0318 00:58:32.655455 2013098752 net.cpp:167] ip1_ip1_0_split needs backward computation.I0318 00:58:32.655462 2013098752 net.cpp:167] ip1 needs backward computation.I0318 00:58:32.655469 2013098752 net.cpp:169] label_data_1_split does not need backward computation.I0318 00:58:32.655477 2013098752 net.cpp:169] data does not need backward computation.I0318 00:58:32.655483 2013098752 net.cpp:205] This network produces output accuracyI0318 00:58:32.655489 2013098752 net.cpp:205] This network produces output lossI0318 00:58:32.655503 2013098752 net.cpp:447] Collecting Learning Rate and Weight Decay.I0318 00:58:32.655511 2013098752 net.cpp:217] Network initialization done.I0318 00:58:32.655517 2013098752 net.cpp:218] Memory required for data: 528I0318 00:58:32.655547 2013098752 solver.cpp:42] Solver scaffolding done.I0318 00:58:32.655567 2013098752 solver.cpp:222] Solving I0318 00:58:32.655575 2013098752 solver.cpp:223] Learning Rate Policy: stepI0318 00:58:32.655583 2013098752 solver.cpp:266] Iteration 0, Testing net (#0)I0318 00:58:32.683643 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.3736I0318 00:58:32.683686 2013098752 solver.cpp:315]     Test net output #1: loss = 1.00555 (* 1 = 1.00555 loss)I0318 00:58:32.683846 2013098752 solver.cpp:189] Iteration 0, loss = 0.869394I0318 00:58:32.683861 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.3I0318 00:58:32.683871 2013098752 solver.cpp:204]     Train net output #1: loss = 0.869394 (* 1 = 0.869394 loss)I0318 00:58:32.683883 2013098752 solver.cpp:464] Iteration 0, lr = 0.01I0318 00:58:32.698721 2013098752 solver.cpp:266] Iteration 1000, Testing net (#0)I0318 00:58:32.701917 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7848I0318 00:58:32.701961 2013098752 solver.cpp:315]     Test net output #1: loss = 0.590972 (* 1 = 0.590972 loss)I0318 00:58:32.702014 2013098752 solver.cpp:189] Iteration 1000, loss = 0.54742I0318 00:58:32.702029 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.7I0318 00:58:32.702041 2013098752 solver.cpp:204]     Train net output #1: loss = 0.54742 (* 1 = 0.54742 loss)I0318 00:58:32.702051 2013098752 solver.cpp:464] Iteration 1000, lr = 0.01I0318 00:58:32.718360 2013098752 solver.cpp:266] Iteration 2000, Testing net (#0)I0318 00:58:32.721529 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7696I0318 00:58:32.721562 2013098752 solver.cpp:315]     Test net output #1: loss = 0.593946 (* 1 = 0.593946 loss)I0318 00:58:32.721593 2013098752 solver.cpp:189] Iteration 2000, loss = 0.729569I0318 00:58:32.721603 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.5I0318 00:58:32.721613 2013098752 solver.cpp:204]     Train net output #1: loss = 0.729569 (* 1 = 0.729569 loss)I0318 00:58:32.721622 2013098752 solver.cpp:464] Iteration 2000, lr = 0.01I0318 00:58:32.740182 2013098752 solver.cpp:266] Iteration 3000, Testing net (#0)I0318 00:58:32.743494 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.77I0318 00:58:32.743544 2013098752 solver.cpp:315]     Test net output #1: loss = 0.591229 (* 1 = 0.591229 loss)I0318 00:58:32.744209 2013098752 solver.cpp:189] Iteration 3000, loss = 0.406097I0318 00:58:32.744231 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.8I0318 00:58:32.744249 2013098752 solver.cpp:204]     Train net output #1: loss = 0.406096 (* 1 = 0.406096 loss)I0318 00:58:32.744266 2013098752 solver.cpp:464] Iteration 3000, lr = 0.01I0318 00:58:32.764135 2013098752 solver.cpp:266] Iteration 4000, Testing net (#0)I0318 00:58:32.769110 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7848I0318 00:58:32.769170 2013098752 solver.cpp:315]     Test net output #1: loss = 0.590972 (* 1 = 0.590972 loss)I0318 00:58:32.769223 2013098752 solver.cpp:189] Iteration 4000, loss = 0.54742I0318 00:58:32.769242 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.7I0318 00:58:32.769255 2013098752 solver.cpp:204]     Train net output #1: loss = 0.54742 (* 1 = 0.54742 loss)I0318 00:58:32.769265 2013098752 solver.cpp:464] Iteration 4000, lr = 0.01I0318 00:58:32.785846 2013098752 solver.cpp:266] Iteration 5000, Testing net (#0)I0318 00:58:32.788722 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7696I0318 00:58:32.788751 2013098752 solver.cpp:315]     Test net output #1: loss = 0.593946 (* 1 = 0.593946 loss)I0318 00:58:32.788811 2013098752 solver.cpp:189] Iteration 5000, loss = 0.72957I0318 00:58:32.788833 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.5I0318 00:58:32.788846 2013098752 solver.cpp:204]     Train net output #1: loss = 0.729569 (* 1 = 0.729569 loss)I0318 00:58:32.788856 2013098752 solver.cpp:464] Iteration 5000, lr = 0.001I0318 00:58:32.804762 2013098752 solver.cpp:266] Iteration 6000, Testing net (#0)I0318 00:58:32.808061 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7856I0318 00:58:32.808112 2013098752 solver.cpp:315]     Test net output #1: loss = 0.59028 (* 1 = 0.59028 loss)I0318 00:58:32.808732 2013098752 solver.cpp:189] Iteration 6000, loss = 0.415444I0318 00:58:32.808753 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.9I0318 00:58:32.808773 2013098752 solver.cpp:204]     Train net output #1: loss = 0.415444 (* 1 = 0.415444 loss)I0318 00:58:32.808786 2013098752 solver.cpp:464] Iteration 6000, lr = 0.001I0318 00:58:32.827118 2013098752 solver.cpp:266] Iteration 7000, Testing net (#0)I0318 00:58:32.831614 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7848I0318 00:58:32.831657 2013098752 solver.cpp:315]     Test net output #1: loss = 0.589454 (* 1 = 0.589454 loss)I0318 00:58:32.831707 2013098752 solver.cpp:189] Iteration 7000, loss = 0.538038I0318 00:58:32.831728 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.8I0318 00:58:32.831745 2013098752 solver.cpp:204]     Train net output #1: loss = 0.538037 (* 1 = 0.538037 loss)I0318 00:58:32.831759 2013098752 solver.cpp:464] Iteration 7000, lr = 0.001I0318 00:58:32.849634 2013098752 solver.cpp:266] Iteration 8000, Testing net (#0)I0318 00:58:32.852712 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7796I0318 00:58:32.852748 2013098752 solver.cpp:315]     Test net output #1: loss = 0.589365 (* 1 = 0.589365 loss)I0318 00:58:32.852792 2013098752 solver.cpp:189] Iteration 8000, loss = 0.684219I0318 00:58:32.852840 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.5I0318 00:58:32.852852 2013098752 solver.cpp:204]     Train net output #1: loss = 0.684219 (* 1 = 0.684219 loss)I0318 00:58:32.852861 2013098752 solver.cpp:464] Iteration 8000, lr = 0.001I0318 00:58:32.868440 2013098752 solver.cpp:266] Iteration 9000, Testing net (#0)I0318 00:58:32.871438 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.7816I0318 00:58:32.871461 2013098752 solver.cpp:315]     Test net output #1: loss = 0.589656 (* 1 = 0.589656 loss)I0318 00:58:32.872109 2013098752 solver.cpp:189] Iteration 9000, loss = 0.421879I0318 00:58:32.872131 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.9I0318 00:58:32.872143 2013098752 solver.cpp:204]     Train net output #1: loss = 0.421879 (* 1 = 0.421879 loss)I0318 00:58:32.872153 2013098752 solver.cpp:464] Iteration 9000, lr = 0.001I0318 00:58:32.889981 2013098752 solver.cpp:334] Snapshotting to examples/hdf5_classification/data/train_iter_10000.caffemodelI0318 00:58:32.890224 2013098752 solver.cpp:342] Snapshotting solver state to examples/hdf5_classification/data/train_iter_10000.solverstateI0318 00:58:32.890362 2013098752 solver.cpp:248] Iteration 10000, loss = 0.538933I0318 00:58:32.890380 2013098752 solver.cpp:266] Iteration 10000, Testing net (#0)I0318 00:58:32.893728 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.782I0318 00:58:32.893757 2013098752 solver.cpp:315]     Test net output #1: loss = 0.589366 (* 1 = 0.589366 loss)I0318 00:58:32.893775 2013098752 solver.cpp:253] Optimization Done.I0318 00:58:32.893786 2013098752 caffe.cpp:134] Optimization Done.

看一下上面的输出或者logreg_auto_train.prototxt，就会发现这个模型是简单的逻辑回归，可以通过引入输入权重和输出权重之间的非线性，使这个模型更好一点，变成两层的，把这个网络放在nonlinear_auto_train.prototxt中。新网络最终的准确率比逻辑回归要高：

from caffe import layers as Lfrom caffe import params as Pdef nonlinear_net(hdf5, batch_size):    # one small nonlinearity, one leap for model kind    n = caffe.NetSpec()    n.data, n.label = L.HDF5Data(batch_size=batch_size, source=hdf5, ntop=2)    # define a hidden layer of dimension 40    n.ip1 = L.InnerProduct(n.data, num_output=40, weight_filler=dict(type='xavier'))    # transform the output through the ReLU (rectified linear) non-linearity    n.relu1 = L.ReLU(n.ip1, in_place=True)    # score the (now non-linear) features    n.ip2 = L.InnerProduct(n.ip1, num_output=2, weight_filler=dict(type='xavier'))    # same accuracy and loss as before    n.accuracy = L.Accuracy(n.ip2, n.label)    n.loss = L.SoftmaxWithLoss(n.ip2, n.label)    return n.to_proto()with open('examples/hdf5_classification/nonlinear_auto_train.prototxt', 'w') as f:    f.write(str(nonlinear_net('examples/hdf5_classification/data/train.txt', 10)))with open('examples/hdf5_classification/nonlinear_auto_test.prototxt', 'w') as f:    f.write(str(nonlinear_net('examples/hdf5_classification/data/test.txt', 10)))

%%timeitcaffe.set_mode_cpu()solver = caffe.get_solver('examples/hdf5_classification/nonlinear_solver.prototxt')solver.solve()accuracy = 0batch_size = solver.test_nets[0].blobs['data'].numtest_iters = int(len(Xt) / batch_size)for i in range(test_iters):    solver.test_nets[0].forward()    accuracy += solver.test_nets[0].blobs['accuracy'].dataaccuracy /= test_itersprint("Accuracy: {:.3f}".format(accuracy))

Accuracy: 0.832
Accuracy: 0.832
Accuracy: 0.832
Accuracy: 0.831
1 loops, best of 3: 386 ms per loop

同样也用命令行看一下模型文件和求解过程：

!./build/tools/caffe train -solver examples/hdf5_classification/nonlinear_solver.prototxt

I0318 00:58:43.336922 2013098752 caffe.cpp:117] Use CPU.I0318 00:58:43.654698 2013098752 caffe.cpp:121] Starting OptimizationI0318 00:58:43.654747 2013098752 solver.cpp:32] Initializing solver from parameters: train_net: "examples/hdf5_classification/nonlinear_auto_train.prototxt"test_net: "examples/hdf5_classification/nonlinear_auto_test.prototxt"test_iter: 250test_interval: 1000base_lr: 0.01display: 1000max_iter: 10000lr_policy: "step"gamma: 0.1momentum: 0.9weight_decay: 0.0005stepsize: 5000snapshot: 10000snapshot_prefix: "examples/hdf5_classification/data/train"solver_mode: CPUI0318 00:58:43.654855 2013098752 solver.cpp:61] Creating training net from train_net file: examples/hdf5_classification/nonlinear_auto_train.prototxtI0318 00:58:43.655004 2013098752 net.cpp:42] Initializing net from parameters: state {  phase: TRAIN}layer {  name: "data"  type: "HDF5Data"  top: "data"  top: "label"  hdf5_data_param {    source: "examples/hdf5_classification/data/train.txt"    batch_size: 10  }}layer {  name: "ip1"  type: "InnerProduct"  bottom: "data"  top: "ip1"  inner_product_param {    num_output: 40    weight_filler {      type: "xavier"    }  }}layer {  name: "relu1"  type: "ReLU"  bottom: "ip1"  top: "ip1"}layer {  name: "ip2"  type: "InnerProduct"  bottom: "ip1"  top: "ip2"  inner_product_param {    num_output: 2    weight_filler {      type: "xavier"    }  }}layer {  name: "accuracy"  type: "Accuracy"  bottom: "ip2"  bottom: "label"  top: "accuracy"}layer {  name: "loss"  type: "SoftmaxWithLoss"  bottom: "ip2"  bottom: "label"  top: "loss"}I0318 00:58:43.655120 2013098752 layer_factory.hpp:74] Creating layer dataI0318 00:58:43.655139 2013098752 net.cpp:84] Creating Layer dataI0318 00:58:43.655264 2013098752 net.cpp:338] data -> dataI0318 00:58:43.655297 2013098752 net.cpp:338] data -> labelI0318 00:58:43.655310 2013098752 net.cpp:113] Setting up dataI0318 00:58:43.655318 2013098752 hdf5_data_layer.cpp:66] Loading list of HDF5 filenames from: examples/hdf5_classification/data/train.txtI0318 00:58:43.655365 2013098752 hdf5_data_layer.cpp:80] Number of HDF5 files: 2I0318 00:58:43.657317 2013098752 net.cpp:120] Top shape: 10 4 (40)I0318 00:58:43.657342 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:43.657356 2013098752 layer_factory.hpp:74] Creating layer label_data_1_splitI0318 00:58:43.657373 2013098752 net.cpp:84] Creating Layer label_data_1_splitI0318 00:58:43.657384 2013098752 net.cpp:380] label_data_1_split <- labelI0318 00:58:43.657395 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_0I0318 00:58:43.657407 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_1I0318 00:58:43.657418 2013098752 net.cpp:113] Setting up label_data_1_splitI0318 00:58:43.657426 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:43.657433 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:43.657441 2013098752 layer_factory.hpp:74] Creating layer ip1I0318 00:58:43.657451 2013098752 net.cpp:84] Creating Layer ip1I0318 00:58:43.657459 2013098752 net.cpp:380] ip1 <- dataI0318 00:58:43.657467 2013098752 net.cpp:338] ip1 -> ip1I0318 00:58:43.657479 2013098752 net.cpp:113] Setting up ip1I0318 00:58:43.662454 2013098752 net.cpp:120] Top shape: 10 40 (400)I0318 00:58:43.662477 2013098752 layer_factory.hpp:74] Creating layer relu1I0318 00:58:43.662497 2013098752 net.cpp:84] Creating Layer relu1I0318 00:58:43.662508 2013098752 net.cpp:380] relu1 <- ip1I0318 00:58:43.662520 2013098752 net.cpp:327] relu1 -> ip1 (in-place)I0318 00:58:43.662530 2013098752 net.cpp:113] Setting up relu1I0318 00:58:43.662539 2013098752 net.cpp:120] Top shape: 10 40 (400)I0318 00:58:43.662546 2013098752 layer_factory.hpp:74] Creating layer ip2I0318 00:58:43.662555 2013098752 net.cpp:84] Creating Layer ip2I0318 00:58:43.662562 2013098752 net.cpp:380] ip2 <- ip1I0318 00:58:43.662571 2013098752 net.cpp:338] ip2 -> ip2I0318 00:58:43.662580 2013098752 net.cpp:113] Setting up ip2I0318 00:58:43.662595 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:43.662606 2013098752 layer_factory.hpp:74] Creating layer ip2_ip2_0_splitI0318 00:58:43.662654 2013098752 net.cpp:84] Creating Layer ip2_ip2_0_splitI0318 00:58:43.662665 2013098752 net.cpp:380] ip2_ip2_0_split <- ip2I0318 00:58:43.662678 2013098752 net.cpp:338] ip2_ip2_0_split -> ip2_ip2_0_split_0I0318 00:58:43.662689 2013098752 net.cpp:338] ip2_ip2_0_split -> ip2_ip2_0_split_1I0318 00:58:43.662698 2013098752 net.cpp:113] Setting up ip2_ip2_0_splitI0318 00:58:43.662706 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:43.662714 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:43.662722 2013098752 layer_factory.hpp:74] Creating layer accuracyI0318 00:58:43.662734 2013098752 net.cpp:84] Creating Layer accuracyI0318 00:58:43.662740 2013098752 net.cpp:380] accuracy <- ip2_ip2_0_split_0I0318 00:58:43.662749 2013098752 net.cpp:380] accuracy <- label_data_1_split_0I0318 00:58:43.662756 2013098752 net.cpp:338] accuracy -> accuracyI0318 00:58:43.662766 2013098752 net.cpp:113] Setting up accuracyI0318 00:58:43.662818 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:43.662827 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:43.662839 2013098752 net.cpp:84] Creating Layer lossI0318 00:58:43.662847 2013098752 net.cpp:380] loss <- ip2_ip2_0_split_1I0318 00:58:43.662854 2013098752 net.cpp:380] loss <- label_data_1_split_1I0318 00:58:43.662863 2013098752 net.cpp:338] loss -> lossI0318 00:58:43.662873 2013098752 net.cpp:113] Setting up lossI0318 00:58:43.662883 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:43.662901 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:43.662909 2013098752 net.cpp:122]     with loss weight 1I0318 00:58:43.662922 2013098752 net.cpp:167] loss needs backward computation.I0318 00:58:43.662930 2013098752 net.cpp:169] accuracy does not need backward computation.I0318 00:58:43.662936 2013098752 net.cpp:167] ip2_ip2_0_split needs backward computation.I0318 00:58:43.662942 2013098752 net.cpp:167] ip2 needs backward computation.I0318 00:58:43.662976 2013098752 net.cpp:167] relu1 needs backward computation.I0318 00:58:43.662988 2013098752 net.cpp:167] ip1 needs backward computation.I0318 00:58:43.662997 2013098752 net.cpp:169] label_data_1_split does not need backward computation.I0318 00:58:43.663003 2013098752 net.cpp:169] data does not need backward computation.I0318 00:58:43.663009 2013098752 net.cpp:205] This network produces output accuracyI0318 00:58:43.663017 2013098752 net.cpp:205] This network produces output lossI0318 00:58:43.663028 2013098752 net.cpp:447] Collecting Learning Rate and Weight Decay.I0318 00:58:43.663035 2013098752 net.cpp:217] Network initialization done.I0318 00:58:43.663041 2013098752 net.cpp:218] Memory required for data: 3728I0318 00:58:43.663158 2013098752 solver.cpp:154] Creating test net (#0) specified by test_net file: examples/hdf5_classification/nonlinear_auto_test.prototxtI0318 00:58:43.663179 2013098752 net.cpp:42] Initializing net from parameters: state {  phase: TEST}layer {  name: "data"  type: "HDF5Data"  top: "data"  top: "label"  hdf5_data_param {    source: "examples/hdf5_classification/data/test.txt"    batch_size: 10  }}layer {  name: "ip1"  type: "InnerProduct"  bottom: "data"  top: "ip1"  inner_product_param {    num_output: 40    weight_filler {      type: "xavier"    }  }}layer {  name: "relu1"  type: "ReLU"  bottom: "ip1"  top: "ip1"}layer {  name: "ip2"  type: "InnerProduct"  bottom: "ip1"  top: "ip2"  inner_product_param {    num_output: 2    weight_filler {      type: "xavier"    }  }}layer {  name: "accuracy"  type: "Accuracy"  bottom: "ip2"  bottom: "label"  top: "accuracy"}layer {  name: "loss"  type: "SoftmaxWithLoss"  bottom: "ip2"  bottom: "label"  top: "loss"}I0318 00:58:43.663349 2013098752 layer_factory.hpp:74] Creating layer dataI0318 00:58:43.663365 2013098752 net.cpp:84] Creating Layer dataI0318 00:58:43.663373 2013098752 net.cpp:338] data -> dataI0318 00:58:43.663385 2013098752 net.cpp:338] data -> labelI0318 00:58:43.663396 2013098752 net.cpp:113] Setting up dataI0318 00:58:43.663422 2013098752 hdf5_data_layer.cpp:66] Loading list of HDF5 filenames from: examples/hdf5_classification/data/test.txtI0318 00:58:43.663457 2013098752 hdf5_data_layer.cpp:80] Number of HDF5 files: 1I0318 00:58:43.664719 2013098752 net.cpp:120] Top shape: 10 4 (40)I0318 00:58:43.664739 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:43.664754 2013098752 layer_factory.hpp:74] Creating layer label_data_1_splitI0318 00:58:43.664772 2013098752 net.cpp:84] Creating Layer label_data_1_splitI0318 00:58:43.664783 2013098752 net.cpp:380] label_data_1_split <- labelI0318 00:58:43.664791 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_0I0318 00:58:43.664803 2013098752 net.cpp:338] label_data_1_split -> label_data_1_split_1I0318 00:58:43.664813 2013098752 net.cpp:113] Setting up label_data_1_splitI0318 00:58:43.664822 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:43.664829 2013098752 net.cpp:120] Top shape: 10 (10)I0318 00:58:43.664837 2013098752 layer_factory.hpp:74] Creating layer ip1I0318 00:58:43.664846 2013098752 net.cpp:84] Creating Layer ip1I0318 00:58:43.664854 2013098752 net.cpp:380] ip1 <- dataI0318 00:58:43.664862 2013098752 net.cpp:338] ip1 -> ip1I0318 00:58:43.664875 2013098752 net.cpp:113] Setting up ip1I0318 00:58:43.664901 2013098752 net.cpp:120] Top shape: 10 40 (400)I0318 00:58:43.664924 2013098752 layer_factory.hpp:74] Creating layer relu1I0318 00:58:43.664945 2013098752 net.cpp:84] Creating Layer relu1I0318 00:58:43.664958 2013098752 net.cpp:380] relu1 <- ip1I0318 00:58:43.664966 2013098752 net.cpp:327] relu1 -> ip1 (in-place)I0318 00:58:43.664975 2013098752 net.cpp:113] Setting up relu1I0318 00:58:43.664983 2013098752 net.cpp:120] Top shape: 10 40 (400)I0318 00:58:43.664990 2013098752 layer_factory.hpp:74] Creating layer ip2I0318 00:58:43.665000 2013098752 net.cpp:84] Creating Layer ip2I0318 00:58:43.665006 2013098752 net.cpp:380] ip2 <- ip1I0318 00:58:43.665015 2013098752 net.cpp:338] ip2 -> ip2I0318 00:58:43.665030 2013098752 net.cpp:113] Setting up ip2I0318 00:58:43.665052 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:43.665066 2013098752 layer_factory.hpp:74] Creating layer ip2_ip2_0_splitI0318 00:58:43.665077 2013098752 net.cpp:84] Creating Layer ip2_ip2_0_splitI0318 00:58:43.665086 2013098752 net.cpp:380] ip2_ip2_0_split <- ip2I0318 00:58:43.665093 2013098752 net.cpp:338] ip2_ip2_0_split -> ip2_ip2_0_split_0I0318 00:58:43.665103 2013098752 net.cpp:338] ip2_ip2_0_split -> ip2_ip2_0_split_1I0318 00:58:43.665113 2013098752 net.cpp:113] Setting up ip2_ip2_0_splitI0318 00:58:43.665122 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:43.665128 2013098752 net.cpp:120] Top shape: 10 2 (20)I0318 00:58:43.665137 2013098752 layer_factory.hpp:74] Creating layer accuracyI0318 00:58:43.665144 2013098752 net.cpp:84] Creating Layer accuracyI0318 00:58:43.665153 2013098752 net.cpp:380] accuracy <- ip2_ip2_0_split_0I0318 00:58:43.665168 2013098752 net.cpp:380] accuracy <- label_data_1_split_0I0318 00:58:43.665180 2013098752 net.cpp:338] accuracy -> accuracyI0318 00:58:43.665192 2013098752 net.cpp:113] Setting up accuracyI0318 00:58:43.665200 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:43.665207 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:43.665216 2013098752 net.cpp:84] Creating Layer lossI0318 00:58:43.665223 2013098752 net.cpp:380] loss <- ip2_ip2_0_split_1I0318 00:58:43.665230 2013098752 net.cpp:380] loss <- label_data_1_split_1I0318 00:58:43.665241 2013098752 net.cpp:338] loss -> lossI0318 00:58:43.665251 2013098752 net.cpp:113] Setting up lossI0318 00:58:43.665259 2013098752 layer_factory.hpp:74] Creating layer lossI0318 00:58:43.665273 2013098752 net.cpp:120] Top shape: (1)I0318 00:58:43.665282 2013098752 net.cpp:122]     with loss weight 1I0318 00:58:43.665290 2013098752 net.cpp:167] loss needs backward computation.I0318 00:58:43.665338 2013098752 net.cpp:169] accuracy does not need backward computation.I0318 00:58:43.665351 2013098752 net.cpp:167] ip2_ip2_0_split needs backward computation.I0318 00:58:43.665380 2013098752 net.cpp:167] ip2 needs backward computation.I0318 00:58:43.665387 2013098752 net.cpp:167] relu1 needs backward computation.I0318 00:58:43.665393 2013098752 net.cpp:167] ip1 needs backward computation.I0318 00:58:43.665400 2013098752 net.cpp:169] label_data_1_split does not need backward computation.I0318 00:58:43.665407 2013098752 net.cpp:169] data does not need backward computation.I0318 00:58:43.665415 2013098752 net.cpp:205] This network produces output accuracyI0318 00:58:43.665421 2013098752 net.cpp:205] This network produces output lossI0318 00:58:43.665431 2013098752 net.cpp:447] Collecting Learning Rate and Weight Decay.I0318 00:58:43.665441 2013098752 net.cpp:217] Network initialization done.I0318 00:58:43.665446 2013098752 net.cpp:218] Memory required for data: 3728I0318 00:58:43.665534 2013098752 solver.cpp:42] Solver scaffolding done.I0318 00:58:43.665568 2013098752 solver.cpp:222] Solving I0318 00:58:43.665577 2013098752 solver.cpp:223] Learning Rate Policy: stepI0318 00:58:43.665586 2013098752 solver.cpp:266] Iteration 0, Testing net (#0)I0318 00:58:43.683938 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.5184I0318 00:58:43.683981 2013098752 solver.cpp:315]     Test net output #1: loss = 0.716141 (* 1 = 0.716141 loss)I0318 00:58:43.684236 2013098752 solver.cpp:189] Iteration 0, loss = 0.764954I0318 00:58:43.684267 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.5I0318 00:58:43.684285 2013098752 solver.cpp:204]     Train net output #1: loss = 0.764954 (* 1 = 0.764954 loss)I0318 00:58:43.684305 2013098752 solver.cpp:464] Iteration 0, lr = 0.01I0318 00:58:43.714700 2013098752 solver.cpp:266] Iteration 1000, Testing net (#0)I0318 00:58:43.721762 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8168I0318 00:58:43.721818 2013098752 solver.cpp:315]     Test net output #1: loss = 0.434918 (* 1 = 0.434918 loss)I0318 00:58:43.721899 2013098752 solver.cpp:189] Iteration 1000, loss = 0.282425I0318 00:58:43.721917 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.9I0318 00:58:43.721932 2013098752 solver.cpp:204]     Train net output #1: loss = 0.282426 (* 1 = 0.282426 loss)I0318 00:58:43.721942 2013098752 solver.cpp:464] Iteration 1000, lr = 0.01I0318 00:58:43.750509 2013098752 solver.cpp:266] Iteration 2000, Testing net (#0)I0318 00:58:43.754590 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8224I0318 00:58:43.754621 2013098752 solver.cpp:315]     Test net output #1: loss = 0.416874 (* 1 = 0.416874 loss)I0318 00:58:43.754660 2013098752 solver.cpp:189] Iteration 2000, loss = 0.51988I0318 00:58:43.754672 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.7I0318 00:58:43.754683 2013098752 solver.cpp:204]     Train net output #1: loss = 0.51988 (* 1 = 0.51988 loss)I0318 00:58:43.754690 2013098752 solver.cpp:464] Iteration 2000, lr = 0.01I0318 00:58:43.782609 2013098752 solver.cpp:266] Iteration 3000, Testing net (#0)I0318 00:58:43.789728 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8176I0318 00:58:43.789777 2013098752 solver.cpp:315]     Test net output #1: loss = 0.415907 (* 1 = 0.415907 loss)I0318 00:58:43.790487 2013098752 solver.cpp:189] Iteration 3000, loss = 0.5093I0318 00:58:43.790510 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.7I0318 00:58:43.790530 2013098752 solver.cpp:204]     Train net output #1: loss = 0.509301 (* 1 = 0.509301 loss)I0318 00:58:43.790544 2013098752 solver.cpp:464] Iteration 3000, lr = 0.01I0318 00:58:43.817451 2013098752 solver.cpp:266] Iteration 4000, Testing net (#0)I0318 00:58:43.821740 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8252I0318 00:58:43.821770 2013098752 solver.cpp:315]     Test net output #1: loss = 0.409124 (* 1 = 0.409124 loss)I0318 00:58:43.821822 2013098752 solver.cpp:189] Iteration 4000, loss = 0.284815I0318 00:58:43.821835 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.9I0318 00:58:43.821846 2013098752 solver.cpp:204]     Train net output #1: loss = 0.284815 (* 1 = 0.284815 loss)I0318 00:58:43.821890 2013098752 solver.cpp:464] Iteration 4000, lr = 0.01I0318 00:58:43.847015 2013098752 solver.cpp:266] Iteration 5000, Testing net (#0)I0318 00:58:43.852102 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8256I0318 00:58:43.852145 2013098752 solver.cpp:315]     Test net output #1: loss = 0.404445 (* 1 = 0.404445 loss)I0318 00:58:43.852188 2013098752 solver.cpp:189] Iteration 5000, loss = 0.511566I0318 00:58:43.852200 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.7I0318 00:58:43.852210 2013098752 solver.cpp:204]     Train net output #1: loss = 0.511566 (* 1 = 0.511566 loss)I0318 00:58:43.852219 2013098752 solver.cpp:464] Iteration 5000, lr = 0.001I0318 00:58:43.876060 2013098752 solver.cpp:266] Iteration 6000, Testing net (#0)I0318 00:58:43.880080 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8328I0318 00:58:43.880105 2013098752 solver.cpp:315]     Test net output #1: loss = 0.396847 (* 1 = 0.396847 loss)I0318 00:58:43.880700 2013098752 solver.cpp:189] Iteration 6000, loss = 0.397858I0318 00:58:43.880718 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.9I0318 00:58:43.880729 2013098752 solver.cpp:204]     Train net output #1: loss = 0.397858 (* 1 = 0.397858 loss)I0318 00:58:43.880738 2013098752 solver.cpp:464] Iteration 6000, lr = 0.001I0318 00:58:43.913795 2013098752 solver.cpp:266] Iteration 7000, Testing net (#0)I0318 00:58:43.917851 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8316I0318 00:58:43.917876 2013098752 solver.cpp:315]     Test net output #1: loss = 0.398135 (* 1 = 0.398135 loss)I0318 00:58:43.917956 2013098752 solver.cpp:189] Iteration 7000, loss = 0.243849I0318 00:58:43.917971 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.9I0318 00:58:43.917989 2013098752 solver.cpp:204]     Train net output #1: loss = 0.243849 (* 1 = 0.243849 loss)I0318 00:58:43.918002 2013098752 solver.cpp:464] Iteration 7000, lr = 0.001I0318 00:58:43.943681 2013098752 solver.cpp:266] Iteration 8000, Testing net (#0)I0318 00:58:43.947589 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8312I0318 00:58:43.947615 2013098752 solver.cpp:315]     Test net output #1: loss = 0.394763 (* 1 = 0.394763 loss)I0318 00:58:43.947651 2013098752 solver.cpp:189] Iteration 8000, loss = 0.513399I0318 00:58:43.947664 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.7I0318 00:58:43.947674 2013098752 solver.cpp:204]     Train net output #1: loss = 0.513399 (* 1 = 0.513399 loss)I0318 00:58:43.947682 2013098752 solver.cpp:464] Iteration 8000, lr = 0.001I0318 00:58:43.973080 2013098752 solver.cpp:266] Iteration 9000, Testing net (#0)I0318 00:58:43.977033 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.834I0318 00:58:43.977056 2013098752 solver.cpp:315]     Test net output #1: loss = 0.395663 (* 1 = 0.395663 loss)I0318 00:58:43.977710 2013098752 solver.cpp:189] Iteration 9000, loss = 0.399341I0318 00:58:43.977735 2013098752 solver.cpp:204]     Train net output #0: accuracy = 0.9I0318 00:58:43.977746 2013098752 solver.cpp:204]     Train net output #1: loss = 0.399342 (* 1 = 0.399342 loss)I0318 00:58:43.977756 2013098752 solver.cpp:464] Iteration 9000, lr = 0.001I0318 00:58:44.003437 2013098752 solver.cpp:334] Snapshotting to examples/hdf5_classification/data/train_iter_10000.caffemodelI0318 00:58:44.003702 2013098752 solver.cpp:342] Snapshotting solver state to examples/hdf5_classification/data/train_iter_10000.solverstateI0318 00:58:44.003850 2013098752 solver.cpp:248] Iteration 10000, loss = 0.244639I0318 00:58:44.003871 2013098752 solver.cpp:266] Iteration 10000, Testing net (#0)I0318 00:58:44.008216 2013098752 solver.cpp:315]     Test net output #0: accuracy = 0.8308I0318 00:58:44.008252 2013098752 solver.cpp:315]     Test net output #1: loss = 0.397291 (* 1 = 0.397291 loss)I0318 00:58:44.008262 2013098752 solver.cpp:253] Optimization Done.I0318 00:58:44.008270 2013098752 caffe.cpp:134] Optimization Done.

最后把这些数据删掉：

# Clean up (comment this out if you want to examine the hdf5_classification/data directory).shutil.rmtree(dirname)