def addf_wide_and_deep_Regression(inputs, addfScore_feature_columns, dnn_feature_columns, dnn_hidden_units):
deep = layers.DenseFeatures(
dnn_feature_columns, name='deep_inputs')(inputs)
for layerno, numnodes in enumerate(dnn_hidden_units):
deep = layers.Dense(numnodes, activation='relu',
name='dnn_{}'.format(layerno+1))(deep)
addf = layers.DenseFeatures(
addfScore_feature_columns, name='addf_inputs')(inputs)
addf_deep = layers.concatenate([deep, addf], name='addf_deep')
output_addf = layers.Dense(
1, activation='relu', name='pred_addf')(addf_deep)
model = Model(inputs, outputs=[output_addf])
model.compile(optimizer='rmsprop', loss='mse', metrics=[
tf.keras.metrics.RootMeanSquaredError(), 'mae', custom_mse])
return model
def create_keras_model(learning_rate=0.001):
wide, deep, inputs = get_wide_deep()
feature_layer_wide = layers.DenseFeatures(wide, name='wide_features')
feature_layer_deep = layers.DenseFeatures(deep, name='deep_features')
wide_model = feature_layer_wide(inputs)
deep_model = layers.Dense(64, activation='relu', name='DNN_layer1')(feature_layer_deep(inputs))
deep_model = layers.Dense(32, activation='relu', name='DNN_layer2')(deep_model)
wide_deep_model = layers.Dense(1, name='weight')(layers.concatenate([wide_model, deep_model]))
model = models.Model(inputs=inputs, outputs=wide_deep_model)
# Compile Keras model
model.compile(loss='mse', optimizer=tf.keras.optimizers.Adam(lr=learning_rate))
return model
def run_algorithm(run_params):
# train_df, train_df_norm, test_df, test_df_norm = data_prep.prepare_data(run_params.training_years, run_params.testing_years, run_params.hour_resolution, True)
data_df, train_mean, train_std = data_prep.prepare_linear_timeseries_data(
run_params)
n = len(data_df)
train_df = data_df[int(n * 0.05):int(n * 0.8)]
test_df = data_df[int(n * 0.8):]
feature_columns = data_prep.add_feature_columns(train_df)
feature_layer = layers.DenseFeatures(feature_columns)
out = DataFrame()
train_df.pop('datetime')
out['datetime'] = test_df.pop('datetime')
out[run_params.predicted_label] = test_df[
run_params.predicted_label] * train_std + train_mean
# out[run_params.predicted_label] = test_df[run_params.predicted_label]
my_model = get_trained_model(train_df, test_df, feature_layer,
run_params.predicted_label, run_params)
print(my_model.trainable_variables)
print("Train mean: {}".format(train_mean))
print("Train std: {}".format(train_std))
test_df.sort_values("index", inplace=True)
test_features = {name: np.array(value) for name, value in test_df.items()}
out[run_params.predicted_label + '_prediction'] = my_model.predict(
test_features) * train_std + train_mean
# out[run_params.predicted_label + '_prediction'] = my_model.predict(test_features)
return out
def build_dnn_model(nbuckets, nnsize, lr):
# input layer is all float except for pickup_datetime which is a string
STRING_COLS = ['pickup_datetime']
NUMERIC_COLS = (set(CSV_COLUMNS) - set([LABEL_COLUMN, 'key']) -
set(STRING_COLS))
inputs = {
colname: layers.Input(name=colname, shape=(), dtype='float32')
for colname in NUMERIC_COLS
}
inputs.update({
colname: layers.Input(name=colname, shape=(), dtype='string')
for colname in STRING_COLS
})
# transforms
transformed, feature_columns = transform(inputs,
NUMERIC_COLS,
STRING_COLS,
nbuckets=nbuckets)
dnn_inputs = layers.DenseFeatures(feature_columns.values())(transformed)
x = dnn_inputs
for layer, nodes in enumerate(nnsize):
x = layers.Dense(nodes, activation='relu', name='h{}'.format(layer))(x)
output = layers.Dense(1, name='fare')(x)
model = models.Model(inputs, output)
lr_optimizer = tf.keras.optimizers.Adam(learning_rate=lr)
model.compile(optimizer=lr_optimizer, loss='mse', metrics=[rmse, 'mse'])
return model
def TrainTestNet(batch_size, train_size, train_batch, valid_size, valid_batch,
test_size, test_batch):
Initialise()
train_df, validation_df, test_df = ProcessData(train_size, train_batch,
valid_size, valid_batch,
test_size, test_batch)
feature_columns = []
# Create a numerical feature column to represent median_income.
for j in ['A_00R','A_00I','A_01R','A_01I','A_02R','A_02I','A_03R','A_03I',\
'A_10R','A_10I','A_11R','A_11I','A_12R','A_12I','A_13R','A_13I',\
'A_20R','A_20I','A_21R','A_21I','A_22R','A_22I','A_23R','A_23I',\
'A_30R','A_30I','A_31R','A_31I','A_32R','A_32I','A_33R','A_33I',\
'B_00R','B_00I','B_01R','B_01I','B_02R','B_02I','B_03R','B_03I',\
'B_10R','B_10I','B_11R','B_11I','B_12R','B_12I','B_13R','B_13I',\
'B_20R','B_20I','B_21R','B_21I','B_22R','B_22I','B_23R','B_23I',\
'B_30R','B_30I','B_31R','B_31I','B_32R','B_32I','B_33R','B_33I']:
a = tf.feature_column.numeric_column(j)
feature_columns.append(a)
# Convert the list of feature columns into a layer that will later be fed into
# the model.
feature_layer = layers.DenseFeatures(feature_columns)
# The following variables are the hyperparameters.
learning_rate = 0.00045
epochs = 70
# Specify the label
label_name = "Fidelity"
# Establish the model's topography.
my_model = create_model(learning_rate, feature_layer)
# Train the model on the normalized training set. We're passing the entire
# normalized training set, but the model will only use the features
# defined by the feature_layer.
epochs, rmse, history = train_model(my_model, train_df, validation_df,
epochs, label_name, batch_size)
#plot_the_loss_curve(epochs, mse)
# After building a model against the training set, test that model
# against the test set.
test_features = {name: np.array(value) for name, value in test_df.items()}
test_label = np.array(test_features.pop(label_name)) # isolate the label
print("\n Evaluate the new model against the test set:")
#his = my_model.evaluate(x = test_features, y = test_label, batch_size=1)
#my_model.save("modelStorage/alpha")
test_features = test_df.drop(["Fidelity"], axis=1)
mae_history = GetTestLosses(test_label, test_features, my_model,
validation_df)
return np.mean(mae_history, np.mean(mae_history))
def data_preprocessing(self):
"""
batch_size = 5 # 예제를 위해 작은 배치 크기를 사용합니다.
train_ds = self.df_to_dataset(self.train, batch_size=batch_size)
val_ds = self.df_to_dataset(self.val, shuffle=False, batch_size=batch_size)
test_ds = self.df_to_dataset(self.test, shuffle=False, batch_size=batch_size)
for feature_batch, label_batch in train_ds.take(1):
print('전체 특성:', list(feature_batch.keys()))
print('나이 특성의 배치:', feature_batch['age'])
print('타깃의 배치:', label_batch)
# 특성 열을 시험해 보기 위해 샘플 배치를 만듭니다.
self.example_batch = next(iter(train_ds))[0]
age = feature_column.numeric_column("age")
self.demo(age)
"""
feature_columns = []
# 수치형 열
for header in [
'age', 'trestbps', 'chol', 'thalach', 'oldpeak', 'slope', 'ca'
]:
feature_columns.append(feature_column.numeric_column(header))
# 버킷형 열
age = feature_column.numeric_column("age")
age_buckets = feature_column.bucketized_column(
age, boundaries=[18, 25, 30, 35, 40, 45, 50, 55, 60, 65])
feature_columns.append(age_buckets)
# 범주형 열
thal = feature_column.categorical_column_with_vocabulary_list(
'thal', ['fixed', 'normal', 'reversible'])
thal_one_hot = feature_column.indicator_column(thal)
feature_columns.append(thal_one_hot)
# 임베딩 열
thal_embedding = feature_column.embedding_column(thal, dimension=8)
feature_columns.append(thal_embedding)
# 교차 특성 열
crossed_feature = feature_column.crossed_column([age_buckets, thal],
hash_bucket_size=1000)
crossed_feature = feature_column.indicator_column(crossed_feature)
feature_columns.append(crossed_feature)
self.feature_layer = layers.DenseFeatures(feature_columns)
batch_size = 32
self.train_ds = self.df_to_dataset(self.train, batch_size=batch_size)
self.val_ds = self.df_to_dataset(self.val,
shuffle=False,
batch_size=batch_size)
self.test_ds = self.df_to_dataset(self.test,
shuffle=False,
batch_size=batch_size)
def create_feature_layer(feature_columns):
# Convert the list of feature columns into a layer that will later be fed into
# the model.
feature_layer = layers.DenseFeatures(feature_columns)
# Print the first 3 and last 3 rows of the feature_layer's output when applied
# to train_df_norm:
#print(feature_layer(dict(train_df_norm)))
return feature_layer
def _get_dense_feature(inputs, feature, shape=(1, )):
"""
Convert an input into a dense numeric feature
NOTE: Can remove this in the future and
pass inputs[feature] directly
"""
feature_col = feature_column.numeric_column(feature, shape=shape)
dense_feature = layers.DenseFeatures(feature_col)(inputs)
return dense_feature
def __init__(self, feature_cols):
super(KerasModel, self).__init__()
self.feature_layer = layers.DenseFeatures(feature_cols)
self.dense_1 = layers.Dense(128, activation='relu')
self.dense_2 = layers.Dense(128, activation='relu')
self.dropout = layers.Dropout(.1)
self.final = layers.Dense(1)
self.compile(optimizer=keras.optimizers.Adam(learning_rate=1e-3),
loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=['accuracy'])
def create_crossed_feature(feature_columns):
latitude = feature_columns[0]
longitude = feature_columns[1]
# Create a feature cross of latitude and longitude.
latitude_x_longitude = tf.feature_column.crossed_column([latitude, longitude], hash_bucket_size=100)
crossed_feature = tf.feature_column.indicator_column(latitude_x_longitude)
feature_columns.append(crossed_feature)
# Convert the list of feature columns into a layer that will later be fed into
# the model.
feature_cross_feature_layer = layers.DenseFeatures(feature_columns)
return feature_cross_feature_layer
def categorical_embedding_with_indices(feature_tensor, feature_info, file_io: FileIO):
"""
Converts input integer tensor into categorical embedding.
Works by converting the categorical indices in the input feature_tensor,
represented as integer values, into categorical embeddings based on the feature_info.
Parameters
----------
feature_tensor : Tensor object
int feature tensor
feature_info : dict
Dictionary representing the configuration parameters for the specific feature from the FeatureConfig
file_io : FileIO object
FileIO handler object for reading and writing
Returns
-------
Tensor object
categorical embedding for the input feature_tensor
Notes
-----
Args under feature_layer_info:
num_buckets : int
Maximum number of categorical values
default_value : int
default value to be assigned to indices out of the num_buckets range
embedding_size : int
dimension size of the categorical embedding
String based categorical features should already be converted into numeric indices
"""
feature_layer_info = feature_info.get("feature_layer_info")
categorical_fc = feature_column.categorical_column_with_identity(
CATEGORICAL_VARIABLE,
num_buckets=feature_layer_info["args"]["num_buckets"],
default_value=feature_layer_info["args"].get("default_value", None),
)
embedding_fc = feature_column.embedding_column(
categorical_fc, dimension=feature_layer_info["args"]["embedding_size"], trainable=True
)
embedding = layers.DenseFeatures(
embedding_fc,
name="{}_embedding".format(feature_info.get("node_name", feature_info["name"])),
)({CATEGORICAL_VARIABLE: feature_tensor})
embedding = tf.expand_dims(embedding, axis=1)
return embedding
def build_tower(features, group, model, params, training):
# group_features = {}
# for feature_name, feature in params["feature_table"].items():
# if group == feature.group:
# group_features[feature_name] = features[feature_name]
columns = params[
"user_context_columns"] if group == "USER_CONTEXT" else params[
"item_columns"]
feature_layer = layers.DenseFeatures(columns)
emb_input = feature_layer(features)
logging.info("%s input dim: %d", group, emb_input.shape[-1])
output = model(emb_input, training)
return output
def _create_model_for_dict_mapping():
model = tf.keras.Sequential()
model.add(
layers.DenseFeatures([
tf.feature_column.numeric_column("a"),
tf.feature_column.numeric_column("b"),
tf.feature_column.numeric_column("c"),
tf.feature_column.numeric_column("d"),
]))
model.add(layers.Dense(16, activation="relu", input_shape=(4, )))
model.add(layers.Dense(3, activation="softmax"))
model.compile(optimizer=tf.keras.optimizers.Adam(),
loss="categorical_crossentropy",
metrics=["accuracy"])
return model
def create_model(_feature_columns):
_feature_layer = layers.DenseFeatures(_feature_columns)
_model = tf.keras.models.Sequential([
_feature_layer,
layers.Dense(100, activation='relu'),
layers.Dense(100, activation='relu'),
layers.Dense(40, activation='softmax')
])
_model.compile(
optimizer='adam',
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
metrics=[tf.metrics.CategoricalCrossentropy()])
return _model
def __init__(self, features):
"""Constructor for classifier.
Args:
features: a list of tf.feature_columns
Returns:
None
"""
super(FairClassifier, self).__init__()
self.d1 = layers.Dense(128, activation='relu', name='dense_1')
self.d2 = layers.Dense(64, activation='relu')
self.d3 = layers.Dense(32, activation='relu')
self.dropout = layers.Dropout(0.2)
self.o = layers.Dense(1, activation='sigmoid')
self.feature_layer = layers.DenseFeatures(features)
def build_dnn_model():
NUM_COLS = [
'pickuplon',
'pickuplat',
'dropofflon',
'dropofflat',
]
CAT_COLS = [
'hourofday',
'dayofweek',
]
inputs = {
colname: layers.Input(name=colname, shape=(), dtype='float32')
for colname in NUM_COLS
}
inputs.update({
colname: layers.Input(name=colname, shape=(), dtype='int32')
for colname in CAT_COLS
})
# transforms
transformed, feature_columns = transform(inputs,
num_cols=NUM_COLS,
cat_cols=CAT_COLS)
dnn_inputs = layers.DenseFeatures(feature_columns.values())(transformed)
# two hidden layers of [32, 8] just in like the BQML DNN
h1 = layers.Dense(32, activation='relu', name='h1')(dnn_inputs)
h2 = layers.Dense(8, activation='relu', name='h2')(h1)
# final output is a linear activation because this is regression
output = layers.Dense(1, activation='linear', name='fare')(h2)
model = models.Model(inputs, output)
# Compile model
model.compile(optimizer='adam',
loss='mse',
metrics=['RootMeanSquaredError'])
return model
def create_bucket_features(train_df, resolution_in_degrees):
'''
Each bin represents all the neighborhoods within a single degree.
For example, neighborhoods at latitude 35.4 and 35.8 are in the same bucket,
but neighborhoods in latitude 35.4 and 36.2 are in different buckets.
The model will learn a separate weight for each bucket.
For example, the model will learn one weight for all the neighborhoods in the "35" bin",
a different weight for neighborhoods in the "36" bin, and so on.
This representation will create approximately 20 buckets:
10 buckets for latitude.
10 buckets for longitude.
'''
# Create a new empty list that will eventually hold the generated feature column.
feature_columns = []
# Create a bucket feature column for latitude.
latitude_as_a_numeric_column = tf.feature_column.numeric_column("latitude")
latitude_boundaries = list(np.arange(int(min(train_df['latitude'])),
int(max(train_df['latitude'])),
resolution_in_degrees))
latitude = tf.feature_column.bucketized_column(latitude_as_a_numeric_column,
latitude_boundaries)
feature_columns.append(latitude)
# Create a bucket feature column for longitude.
longitude_as_a_numeric_column = tf.feature_column.numeric_column("longitude")
longitude_boundaries = list(np.arange(int(min(train_df['longitude'])),
int(max(train_df['longitude'])),
resolution_in_degrees))
longitude = tf.feature_column.bucketized_column(longitude_as_a_numeric_column,
longitude_boundaries)
feature_columns.append(longitude)
# Convert the list of feature columns into a layer that will ultimately become
# part of the model. Understanding layers is not important right now.
buckets_feature_layer = layers.DenseFeatures(feature_columns)
return feature_columns, buckets_feature_layer
def create_model(my_learning_rate,
feature_column_names,
neural_network_structure,
dropout_rate=0.2):
feature_columns = []
for feature_column_name in feature_column_names:
feature = tf.feature_column.numeric_column(feature_column_name)
feature_columns.append(feature)
feature_layer = layers.DenseFeatures(feature_columns)
model = tf.keras.models.Sequential()
model.add(feature_layer)
model.add(layers.BatchNormalization())
index = 0
for number_of_nodes in neural_network_structure:
index = index + 1
if index == 2 and dropout_rate > 0:
print('adding dropout layer', dropout_rate)
model.add(tf.keras.layers.Dropout(rate=dropout_rate))
layer_name = 'Hidden' + str(index)
print('adding hidden layer', layer_name, 'with', number_of_nodes,
'nodes')
model.add(
tf.keras.layers.Dense(
units=number_of_nodes,
activation='relu',
kernel_regularizer=tf.keras.regularizers.l2(l=0.01),
name=layer_name))
model.add(tf.keras.layers.Dense(units=1, name='Output'))
model.compile(optimizer=tf.keras.optimizers.RMSprop(lr=my_learning_rate),
loss="mean_squared_error",
metrics=[tf.keras.metrics.RootMeanSquaredError()])
return model
def do_research(run_params):
output_file_path = 'C:\\Users\\sliwk\\Downloads\\Dane\\Output\\' + run_params.hour_resolution + 'h out.csv'
file = open(output_file_path, "a")
file.close()
train_df, train_df_norm, test_df, test_df_norm = data_prep.prepare_data(
run_params.training_years, run_params.testing_years,
run_params.hour_resolution, True)
feature_columns = data_prep.add_feature_columns(train_df)
feature_layer = layers.DenseFeatures(feature_columns)
my_model = get_trained_model(train_df, test_df, feature_layer,
run_params.predicted_label, run_params)
research_df = data_prep.prepare_research_data(test_df)
counter = 0
for record in research_df.index:
if counter < 2:
counter = counter + 1
continue
data_prep.add_past_one_record(counter, research_df,
run_params.hour_resolution)
lol = research_df.loc[counter:counter]
test_features = {name: np.array(value) for name, value in lol.items()}
prediction = my_model.predict(test_features)
value = prediction.item(0)
research_df.loc[counter, run_params.predicted_label] = value
counter = counter + 1
research_df.sort_values("index", inplace=True)
return research_df
def model(train, val, test, columns):
"Building, compiling, fitting, and evaluating model"
feature_columns = []
for header in columns[0:len(columns) - 1]:
feature_columns.append(feature_column.numeric_column(header))
feature_layer = layers.DenseFeatures(feature_columns)
model = tf.keras.Sequential([
feature_layer,
layers.Dense(64, activation='relu'),
layers.Dropout(0.5),
layers.Dense(64, activation='relu'),
layers.Dropout(0.5),
layers.Dense(1, activation='sigmoid')
])
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
model.fit(train, validation_data=val, epochs=7)
loss, accuracy = model.evaluate(test)
print("Accuracy", accuracy)
def __init__(self,
cross_layer_num,
deep_layer_num,
deep_layer_dim,
batch_size,
feature_columns,
embed_dim,
num_classes=10,
**kwargs):
super(CrossDeep, self).__init__(name='CrossDeep')
self.cross_layer_num = cross_layer_num
self.dense_feature = layers.DenseFeatures(feature_columns)
self.deep_layer_num = deep_layer_num
self.batch_size = batch_size
self.num_classes = num_classes
self.deep_layer_dim = deep_layer_dim
self.embed_dim = embed_dim
self.inputs_shape = 5 * self.embed_dim + 1
self.W = []
self.bias = []
self.dense_list = []
# self.dense = layers.Dense(self.num_classes,activation='sigmoid')
self.softmax = layers.Softmax(input_shape=(449, ))
for i in range(self.deep_layer_num):
self.dense_list.append(
layers.Dense(self.deep_layer_dim, activation='relu'))
b_init = tf.zeros_initializer()
w_init = tf.random_normal_initializer()
for i in range(self.cross_layer_num):
self.W.append(
tf.Variable(initial_value=w_init(shape=(self.batch_size,
self.inputs_shape)),
trainable=True))
self.bias.append(
tf.Variable(initial_value=b_init(shape=(self.batch_size, 1)),
trainable=True))
def categorical_indicator_with_vocabulary_file(feature_tensor, feature_info,
file_io: FileIO):
"""
Converts a string tensor into a categorical one-hot representation.
Works by using a vocabulary file to convert the string tensor into categorical indices
and then converting the categories into one-hot representation.
Args:
feature_tensor: String feature tensor
feature_info: Dictionary representing the configuration parameters for the specific feature from the FeatureConfig
Returns:
Categorical one-hot representation of input feature_tensor
Args under feature_layer_info:
vocabulary_file: string; path to vocabulary CSV file for the input tensor containing the vocabulary to look-up.
uses the "key" named column as vocabulary of the 1st column if no "key" column present.
max_length: int; max number of rows to consider from the vocabulary file.
if null, considers the entire file vocabulary.
num_oov_buckets: int - optional; number of out of vocabulary buckets/slots to be used to
encode strings into categorical indices. If not specified, the default is 1.
NOTE:
The vocabulary CSV file must contain two columns - key, id,
where the key is mapped to one id thereby resulting in a
many-to-one vocabulary mapping.
If id field is absent, a unique whole number id is assigned by default
resulting in a one-to-one mapping
"""
#
##########################################################################
#
# NOTE:
# Current bug[1] with saving a Keras model when using
# feature_column.categorical_column_with_vocabulary_list.
# Tracking the issue currently and should be able to upgrade
# to current latest stable release 2.2.0 to test.
#
# Can not use TF2.1.0 due to issue[2] regarding saving Keras models with
# custom loss, metric layers
#
# Can not use TF2.2.0 due to issues[3, 4] regarding incompatibility of
# Keras Functional API models and Tensorflow
#
# References:
# [1] https://github.com/tensorflow/tensorflow/issues/31686
# [2] https://github.com/tensorflow/tensorflow/issues/36954
# [3] https://github.com/tensorflow/probability/issues/519
# [4] https://github.com/tensorflow/tensorflow/issues/35138
#
# CATEGORICAL_VARIABLE = "categorical_variable"
# categorical_fc = feature_column.categorical_column_with_vocabulary_list(
# CATEGORICAL_VARIABLE,
# vocabulary_list=vocabulary_list,
# default_value=feature_layer_info["args"].get("default_value", -1),
# num_oov_buckets=feature_layer_info["args"].get("num_oov_buckets", 0),
# )
#
# indicator_fc = feature_column.indicator_column(categorical_fc)
#
# categorical_one_hot = layers.DenseFeatures(
# indicator_fc,
# name="{}_one_hot".format(feature_info.get("node_name", feature_info["name"])),
# )({CATEGORICAL_VARIABLE: feature_tensor})
# categorical_one_hot = tf.expand_dims(categorical_one_hot, axis=1)
#
##########################################################################
#
feature_tensor_indices, vocabulary_keys, num_oov_buckets = categorical_indices_from_vocabulary_file(
feature_info, feature_tensor, file_io)
vocabulary_size = len(set(vocabulary_keys))
categorical_identity_fc = feature_column.categorical_column_with_identity(
CATEGORICAL_VARIABLE, num_buckets=vocabulary_size + num_oov_buckets)
indicator_fc = feature_column.indicator_column(categorical_identity_fc)
categorical_one_hot = layers.DenseFeatures(
indicator_fc,
name="{}_one_hot".format(
feature_info.get("node_name", feature_info["name"])),
)({
CATEGORICAL_VARIABLE: feature_tensor_indices
})
categorical_one_hot = tf.expand_dims(categorical_one_hot, axis=1)
return categorical_one_hot
def transform_output(self, featureColumn):
feature_layer = layers.DenseFeatures(featureColumn)
example = next(iter(self.exampleData))[0]
log(feature_layer(example).numpy())
input_fn=lambda: input_fun(test_data, label_column, training=False))
print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result))
#建立一個keras model 來比較運算時間
feature_column_list = []
for feature in feature_column:
feature_column_list.append(tf.feature_column.numeric_column(key=feature))
#tf.data.experimental.make_csv_dataset 的輸出分為兩部分
#第一個為以各個輸入column name為 key的字典
#第二個是包含數筆輸入資料類別的 eager_tensor
#需要用DenseDeature 來取的資料
#
model = Sequential([
layers.DenseFeatures(feature_column_list),
layers.Dense(5, 'relu'),
layers.Dense(3, 'relu'),
layers.Dense(3, 'softmax')
])
input_layer_dict = {}
for feature in feature_column:
input_layer_dict[feature] = Input(shape=(1, ), name=feature)
input_layer = layers.DenseFeatures(feature_column_list)(input_layer_dict)
layer = layers.Dense(5, 'relu')(input_layer)
layer = layers.Dense(3, 'relu')(layer)
output_layer = layers.Dense(3, 'softmax')(layer)
model = Model(input_layer_dict, output_layer)
# Create a feature cross of latitude and longitude.
latitude_x_longitude = tf.feature_column.crossed_column([latitude, longitude],
hash_bucket_size=200)
crossed_feature = tf.feature_column.indicator_column(latitude_x_longitude)
feature_columns.append(crossed_feature)
#Turn rest of useful columns into features and append to feature_columns
day_diff_numeric_column = tf.feature_column.numeric_column("DAY_DIFF")
feature_columns.append(day_diff_numeric_column)
fire_size_numeric_column = tf.feature_column.numeric_column("FIRE_SIZE")
feature_columns.append(fire_size_numeric_column)
# Convert the list of feature columns into a layer that will later be fed into
# the model.
features_layer = layers.DenseFeatures(feature_columns)
def create_model(my_learning_rate, my_feature_layer):
"""Create and compile a deep neural net."""
model = tf.keras.models.Sequential()
# model.add(my_feature_layer)
model.add(
tf.keras.layers.Dense(units=100, activation='relu', input_shape=(4, )))
model.add(tf.keras.layers.Dropout(rate=0.2))
model.add(tf.keras.layers.Dense(units=50, activation='relu'))
model.add(tf.keras.layers.Dropout(rate=0.1))
model.add(tf.keras.layers.Dense(units=13, activation='softmax'))
model.compile(loss=loss_function,
optimizer=optimizer,
def demo_numeric(feature_column):
global a
feature_layer = layers.DenseFeatures(feature_column)
print(feature_layer(a).numpy())
def demo_emb(feature_column):
global c
feature_layer = layers.DenseFeatures(feature_column)
print(feature_layer(c).numpy())
longitude_boundaries = list(
np.arange(int(min(train_df['longitude'])), int(max(train_df['longitude'])),
resolution_in_degrees))
longitude = tf.feature_column.bucketized_column(longitude_as_a_numeric_column,
longitude_boundaries)
feature_columns.append(longitude)
# Create a feature cross of latitude and longitude.
latitude_x_longitude = tf.feature_column.crossed_column([latitude, longitude],
hash_bucket_size=100)
crossed_feature = tf.feature_column.indicator_column(latitude_x_longitude)
feature_columns.append(crossed_feature)
# Convert the list of feature columns into a layer that will ultimately become
# part of the model. Understanding layers is not important right now.
feature_cross_feature_layer = layers.DenseFeatures(feature_columns)
# The following variables are the hyperparameters.
learning_rate = 0.04
epochs = 35
batch_size = 100
label_name = 'median_house_value'
# Build the model, this time passing in the feature_cross_feature_layer.
my_model = create_model(learning_rate, feature_cross_feature_layer)
# Train the model on the training set.
epochs, rmse = train_model(my_model, train_df, epochs, batch_size, label_name)
plot_the_loss_curve(epochs, rmse)
def demo(feature_column):
feature_layer = layers.DenseFeatures(feature_column)
print(feature_layer(example_batch).numpy())
def demo_cat(feature_column):
global c
feature_layer = layers.DenseFeatures(
[tf.feature_column.indicator_column(feature_column)])
print(feature_layer(c).numpy())
