鸢尾花数据集
鸢尾花数据集包含四个特征和一个标签。这四个特征确定了单株鸢尾花的下列植物学特征：花萼长度、花萼宽度、花瓣长度、花瓣宽度。我们的模型会将这些特征表示为float32数值数据。 该标签确定了鸢尾花品种，品种必须是下列任意一种：山鸢尾 (0)、变色鸢尾 (1)、维吉尼亚鸢尾 (2)。我们的模型会将该标签表示为int32分类数据。
算法
该程序会训练一个具有以下拓扑结构的深度神经网络分类器模型：2 个隐藏层，每个隐藏层包含 10 个节点。
下图展示了特征、隐藏层和预测（并未显示隐藏层中的所有节点）：
推理
在无标签样本上运行经过训练的模型会产生三个预测，即相应鸢尾花属于指定品种的可能性。这些输出预测的总和是 1.0。例如，对无标签样本的预测可能如下所示：0.03（山鸢尾）、0.95（变色鸢尾）、0.02（维吉尼亚鸢尾）。该预测表示指定无标签样本是变色鸢尾的概率为 95％。
数据分析
所有功能的出发点都是数据，下面对数据进行全面的分析。首先，这个数据集是网上公开的一个CSV格式的数据集，可以将数据下载，使用pandas对数据进行分析。在此处已经将数据下载好并放在data文件下。
下面使用pandas对数据进行读取，并部分展示，由于下载的数据自带格式所以做了一下处理。
import pandas as pd
import tensorflow as tf
CSV_COLUMN_NAMES = ['SepalLength','SepalWidth','PetalLength', 'PetalWidth', 'Species']
data_train=pd.read_csv('./data/iris_test.csv',names=CSV_COLUMN_NAMES,header=0)
data_test=pd.read_csv('./data/iris_training.csv',names=CSV_COLUMN_NAMES,header=0)
data_train.head()Out[1]:
| SepalLength | SepalWidth | PetalLength | PetalWidth | Species | |
|---|---|---|---|---|---|
| 0 | 5.9 | 3.0 | 4.2 | 1.5 | 1 |
| 1 | 6.9 | 3.1 | 5.4 | 2.1 | 2 |
| 2 | 5.1 | 3.3 | 1.7 | 0.5 | 0 |
| 3 | 6.0 | 3.4 | 4.5 | 1.6 | 1 |
| 4 | 5.5 | 2.5 | 4.0 | 1.3 | 1 |
构建模型
在构建模型结构时暂时不需要数据的输入，但需要制定有几个特征需要输入，从上面的结构中可以看出。输入的特征有四个：花萼长度（seqpallength），花萼宽度（sepalwidth），花瓣长度 (petallength)，花瓣宽度 (petalwidth)。而最后一个是预测值，species 代表他是哪种鸢尾花：山鸢尾(0)，变色鸢尾(1)，维吉尼亚鸢尾 (2)。所以我们先将属性值与标记值区分开
train_x, train_y = data_train, data_train.pop('Species')
test_x, test_y = data_test, data_test.pop('Species')
train_x.head()
Out[2]:
| SepalLength | SepalWidth | PetalLength | PetalWidth | |
|---|---|---|---|---|
| 0 | 5.9 | 3.0 | 4.2 | 1.5 |
| 1 | 6.9 | 3.1 | 5.4 | 2.1 |
| 2 | 5.1 | 3.3 | 1.7 | 0.5 |
| 3 | 6.0 | 3.4 | 4.5 | 1.6 |
| 4 | 5.5 | 2.5 | 4.0 | 1.3 |
my_feature_columns = [] #从train_x提取特征值
for key in train_x.keys():
my_feature_columns.append(tf.feature_column.numeric_column(key=key))
print(my_feature_columns)
[_NumericColumn(key='SepalLength', shape=(1,), default_value=None, dtype=tf.float32, normalizer_fn=None), _NumericColumn(key='SepalWidth', shape=(1,), default_value=None, dtype=tf.float32, normalizer_fn=None), _NumericColumn(key='PetalLength', shape=(1,), default_value=None, dtype=tf.float32, normalizer_fn=None), _NumericColumn(key='PetalWidth', shape=(1,), default_value=None, dtype=tf.float32, normalizer_fn=None)]
下面开始构建训练模型
classifier = tf.estimator.DNNClassifier(
# 这个模型接受哪些输入的特征
feature_columns=my_feature_columns,
# 包含两个隐藏层，每个隐藏层包含10个神经元.
hidden_units=[10, 10],
# 最终结果要分成的几类
n_classes=3)
INFO:tensorflow:Using default config.
WARNING:tensorflow:Using temporary folder as model directory: /tmp/tmpg62ivkem
INFO:tensorflow:Using config: {'_model_dir': '/tmp/tmpg62ivkem', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': allow_soft_placement: true
graph_options {
rewrite_options {
meta_optimizer_iterations: ONE
}
}
, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_service': None, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7efc2dc54160>, '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}
训练模型
下面就需要为模型提供数据并进行训练,由于tensorflow采用一个批量梯度下降算法更新参数，这里可以构造一个函数来生成数据，并且可以在这个函数当中对数据进行打乱。
def train_func(train_x,train_y):
dataset=tf.data.Dataset.from_tensor_slices((dict(train_x), train_y))
dataset = dataset.shuffle(1000).repeat().batch(100)
return dataset下面可以进行模型训练，进行1000个回合的训练，每次调100的数据。
classifier.train(
input_fn=lambda:train_func(train_x,train_y),
steps=1000)
INFO:tensorflow:Calling model_fn. INFO:tensorflow:Done calling model_fn. INFO:tensorflow:Create CheckpointSaverHook. INFO:tensorflow:Graph was finalized. INFO:tensorflow:Running local_init_op. INFO:tensorflow:Done running local_init_op. INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmpg62ivkem/model.ckpt. INFO:tensorflow:loss = 131.49458, step = 0 INFO:tensorflow:global_step/sec: 230.342 INFO:tensorflow:loss = 11.232605, step = 100 (0.435 sec) INFO:tensorflow:global_step/sec: 508.696 INFO:tensorflow:loss = 8.845632, step = 200 (0.196 sec) INFO:tensorflow:global_step/sec: 509.375 INFO:tensorflow:loss = 7.626791, step = 300 (0.196 sec) INFO:tensorflow:global_step/sec: 515.75 INFO:tensorflow:loss = 6.2779765, step = 400 (0.194 sec) INFO:tensorflow:global_step/sec: 459.292 INFO:tensorflow:loss = 6.688465, step = 500 (0.218 sec) INFO:tensorflow:global_step/sec: 509.881 INFO:tensorflow:loss = 6.19946, step = 600 (0.196 sec) INFO:tensorflow:global_step/sec: 523.267 INFO:tensorflow:loss = 6.8552265, step = 700 (0.191 sec) INFO:tensorflow:global_step/sec: 526.077 INFO:tensorflow:loss = 6.5732956, step = 800 (0.190 sec) INFO:tensorflow:global_step/sec: 519.665 INFO:tensorflow:loss = 6.094924, step = 900 (0.192 sec) INFO:tensorflow:Saving checkpoints for 1000 into /tmp/tmpg62ivkem/model.ckpt. INFO:tensorflow:Loss for final step: 5.365243.
Out[6]:
<tensorflow.python.estimator.canned.dnn.DNNClassifier at 0x7efc2dc54b00>
模型预测
可以使用下面方法对测试集的数据进行预测，并查看效果
def eval_input_fn(features, labels, batch_size):
features=dict(features)
if labels is None:
# No labels, use only features.
inputs = features
else:
inputs = (features, labels)
dataset = tf.data.Dataset.from_tensor_slices(inputs)
assert batch_size is not None, "batch_size must not be None"
dataset = dataset.batch(batch_size)
return datasetpredict_arr = []
predictions = classifier.predict(
input_fn=lambda:eval_input_fn(test_x,labels=test_y,batch_size=100))
for predict in predictions:
predict_arr.append(predict['probabilities'].argmax())
result = predict_arr == test_y
result1 = [w for w in result if w == True]
print("准确率为 %s"%str((len(result1)/len(result))))INFO:tensorflow:Calling model_fn. INFO:tensorflow:Done calling model_fn. INFO:tensorflow:Graph was finalized. INFO:tensorflow:Restoring parameters from /tmp/tmpg62ivkem/model.ckpt-1000 INFO:tensorflow:Running local_init_op. INFO:tensorflow:Done running local_init_op. 准确率为 0.9833333333333333