我有两层LSTM网络。 (config.n_input为3,config.n_steps为5)
我认为这可能与我输入的形状有关,但我不确定如何修复它,我尝试更改LSTM的投影,以便它们的输入大小相同,但是没有工作
tensorflow.python.pywrap_tensorflow.StatusNotOK: Invalid argument: Expected size[1] in [0, 0], but got 600
[[Node: RNN/MultiRNNCell/Cell1/Slice = Slice[Index=DT_INT32, T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/gpu:0"](_recv_state_0/_3, RNN/MultiRNNCell/Cell1/Slice/begin, RNN/MultiRNNCell/Cell1/Slice/size)]]
尝试运行时会返回以下错误。
(function() {
angular.module('myApp', []);
var uniqueItems = function(data, key) {
var result = [];
for (var i = 0; i < data.length; i++) {
var value = data[i][key];
if (result.indexOf(value) == -1) {
result.push(value);
}
}
return result;
};
angular.module('myApp').controller("MyCtrl",MyCtrl)
function MyCtrl($scope, filterFilter) {
$scope.usePants = {};
$scope.useShirts = {};
$scope.useShoes = {};
$scope.players = [{
name: 'Bruce Wayne',
shirt: 'XXL',
pants: '42',
shoes: '12'
}, {
name: 'Wayne Gretzky',
shirt: 'XL',
pants: '38',
shoes: '10'
}, {
name: 'Michael Jordan',
shirt: 'M',
pants: '32',
shoes: '9'
}, {
name: 'Rodman',
shirt: 'XSXL',
pants: '42',
shoes: '11'
}, {
name: 'Jake Smitz',
shirt: 'XXL',
pants: '42',
shoes: '12'
}, {
name: 'Will Will',
shirt: 'XXLL',
pants: '42',
shoes: '12'
}, {
name: 'Youasdf Oukls',
shirt: 'XL',
pants: '38',
shoes: '10'
}, {
name: 'Sam Sneed',
shirt: 'XL',
pants: '38',
shoes: '10'
}, {
name: 'Bill Waxy',
shirt: 'M',
pants: '32',
shoes: '9'
}, {
name: 'Javier Xavior',
shirt: 'M',
pants: '32',
shoes: '9'
}, {
name: 'Bill Knight',
shirt: 'M',
pants: '32',
shoes: '9'
}, {
name: 'One More',
shirt: 'M',
pants: '32',
shoes: '9'
}, {
name: 'Player One',
shirt: 'XXL',
pants: '42',
shoes: '11'
}, {
name: 'Space Cadet',
shirt: 'XXL',
pants: '42',
shoes: '12'
}, {
name: 'Player Two',
shirt: 'XXXXL',
pants: '42',
shoes: '12'
} {
name: 'Bill Knight',
shirt: 'M',
pants: '32',
shoes: '9'
}, {
name: 'One More',
shirt: 'M',
pants: '32',
shoes: '9'
}, {
name: 'Player One',
shirt: 'XXL',
pants: '42',
shoes: '11'
}, {
name: 'Space Cadet',
shirt: 'XXL',
pants: '42',
shoes: '12'
}];
// Watch the pants that are selected
$scope.$watch(function() {
return {
players: $scope.players,
usePants: $scope.usePants,
useShirts: $scope.useShirts,
useShoes: $scope.useShoes
}
}, function(value) {
var selected;
$scope.count = function(prop, value) {
return function(el) {
return el[prop] == value;
};
};
$scope.pantsGroup = uniqueItems($scope.players, 'pants');
var filterAfterPants = [];
selected = false;
for (var j in $scope.players) {
var p = $scope.players[j];
for (var i in $scope.usePants) {
if ($scope.usePants[i]) {
selected = true;
if (i == p.pants) {
filterAfterPants.push(p);
break;
}
}
}
}
if (!selected) {
filterAfterPants = $scope.players;
}
$scope.shirtsGroup = uniqueItems($scope.players, 'shirt');
var filterAfterShirts = [];
selected = false;
for (var j in filterAfterPants) {
var p = filterAfterPants[j];
for (var i in $scope.useShirts) {
if ($scope.useShirts[i]) {
selected = true;
if (i == p.shirt) {
filterAfterShirts.push(p);
break;
}
}
}
}
if (!selected) {
filterAfterShirts = filterAfterPants;
}
$scope.shoesGroup = uniqueItems($scope.players, 'shoes');
var filterAfterShoes = [];
selected = false;
for (var j in filterAfterShirts) {
var p = filterAfterShirts[j];
for (var i in $scope.useShoes) {
if ($scope.useShoes[i]) {
selected = true;
if (i == p.shoes) {
filterAfterShoes.push(p);
break;
}
}
}
}
if (!selected) {
filterAfterShoes = filterAfterShirts;
}
$scope.filteredPlayers = filterAfterShoes;
}, true);
$scope.$watch('filtered', function(newValue) {
if (angular.isArray(newValue)) {
console.log(newValue.length);
}
}, true);
}
angular.module('myApp').filter('count', function() {
return function(collection, key) {
var out = "test";
for (var i = 0; i < collection.length; i++) {
//console.log(collection[i].pants);
//var out = myApp.filter('filter')(collection[i].pants, "42", true);
}
return out;
}
});
angular.module('myApp').filter('groupBy',
function() {
return function(collection, key) {
if (collection === null) return;
return uniqueItems(collection, key);
};
});
})();
错误消息的任何优秀解释?或者有什么方法可以轻松解决这个问题吗?
答案 0 :(得分:0)
将num_proj添加到初始状态:
# Tensorflow LSTM cell requires 2x n_hidden length (state & cell)
self.initial_state = tf.placeholder(tf.float32, [None, 2*config.n_hidden + 300], name='state')
这是一个非常不透明的错误,您可以在TF GitHub问题页面上提出它,这可能是一个好主意!