"comment":"\n Creates a print op that will print when a tensor is accessed.\n\n Wraps the tensor passed in so that whenever that a tensor is accessed,\n the message `message` is printed, along with the current value of the\n tensor `t`.",
"inputs":[
{
"name":"input",
"comment":"the tensor that will be displayed.",
"duplicable":0,
"intermediate":0
}],
"outputs":[],
"attrs":[
{
"name":"first_n",
"type":"int",
"comment":"Only log `first_n` number of times.",
"generated":0
},{
"name":"message",
"type":"string",
"comment":"A string message to print as a prefix.",
"generated":0
},{
"name":"summarize",
"type":"int",
"comment":"Print this number of elements in the tensor.",
"generated":0
},{
"name":"print_tensor_name",
"type":"bool",
"comment":"Whether to print the tensor name.",
"generated":0
},{
"name":"print_tensor_type",
"type":"bool",
"comment":"Whether to print the tensor's dtype.",
"generated":0
},{
"name":"print_tensor_shape",
"type":"bool",
"comment":"Whether to print the tensor's shape.",
"generated":0
},{
"name":"print_tensor_lod",
"type":"bool",
"comment":"Whether to print the tensor's lod.",
"generated":0
}]
},{
"type":"adagrad",
"comment":"\n\nAdaptive Gradient Algorithm (Adagrad).\n\nThe update is done as follows:\n\n$$moment\\_out = moment + grad * grad \\\\\nparam\\_out = param - \\frac{learning\\_rate * grad}{\\sqrt{moment\\_out} + \\epsilon}\n$$\n\nThe original paper(http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf)\ndoes not have the epsilon attribute. It is added here in our implementation\nas also proposed here: http://cs231n.github.io/neural-networks-3/#ada\nfor numerical stability to avoid the division by zero error.\n\n",
