public TFFunction(TensorFlowBackend k, List <Tensor> inputs, List <Tensor> outputs, List <List <Tensor> > updates, string name)
{
this.K = k;
this.tf = k.tf;
if (updates == null)
{
updates = new List <List <Tensor> >();
}
this.inputs = inputs;
this.outputs = outputs;
{
var updates_ops = new List <TFOutput>();
foreach (List <Tensor> update in updates)
{
if (update.Count == 2)
{
var p = K.In(update[0]);
var new_p = K.In(update[1]);
updates_ops.Add(tf.Assign(p, new_p));
}
else
{
// assumed already an op
updates_ops.Add(K.In(update[0]));
}
}
//this.updates_op = tf.group(updates_ops);
this.updates_op = updates_ops;
}
this.name = name;
//this.session_kwargs = session_kwargs;
}
public override List <Tensor> Call(List <Array> inputs)
{
var feed_dict = new Dictionary <Tensor, Array>();
foreach (var(tensor, value) in Enumerable.Zip(this.inputs, inputs, (a, b) => (a, b)))
{
// if (is_sparse(tensor))
// {
// sparse_coo = value.tocoo()
// indices = np.concatenate((np.expand_dims(sparse_coo.row, 1),
// np.expand_dims(sparse_coo.col, 1)), 1)
// value = (indices, sparse_coo.data, sparse_coo.shape)
// }
feed_dict[tensor] = value;
}
var session = K._SESSION;
var outputs = new List <TFOutput>();
foreach (var o in this.outputs)
{
outputs.Add(K.In(o));
}
foreach (TFOutput o in this.updates_op)
{
outputs.Add(o);
}
var _inputs = new List <TFOutput>();
var _values = new List <TFTensor>();
foreach (KeyValuePair <Tensor, Array> pair in feed_dict)
{
_inputs.Add(K.In(pair.Key));
_values.Add(pair.Value);
}
var updated = session.Run(
inputs: _inputs.ToArray(),
inputValues: _values.ToArray(),
outputs: outputs.ToArray());
return(updated.Get(0, this.outputs.Count).Select(t => K.Out(t)).ToList());
}
public override List <Tensor> Call(List <Array> inputs)
{
var feed_dict = new Dictionary <Tensor, Array>();
foreach (var(tensor, value) in Enumerable.Zip(this.inputs, inputs, (a, b) => (a, b)))
{
// if (is_sparse(tensor))
// {
// sparse_coo = value.tocoo()
// indices = np.concatenate((np.expand_dims(sparse_coo.row, 1),
// np.expand_dims(sparse_coo.col, 1)), 1)
// value = (indices, sparse_coo.data, sparse_coo.shape)
// }
feed_dict[tensor] = value;
}
var session = K._SESSION;
var init = tf.GetGlobalVariablesInitializer();
if (init.Length > 0)
{
Console.WriteLine("Initializing variables:");
foreach (var op in init)
{
Console.WriteLine(" - " + op.Name);
session.Run(new TFOutput[0], new TFTensor[0], new TFOutput[0], new[] { op });
}
Console.WriteLine("Operations:");
foreach (var op in tf.GetEnumerator())
{
Console.WriteLine(" - " + op.Name);
}
Console.WriteLine();
}
//Console.WriteLine("Before:");
//PrintVariables(feed_dict, session);
// Console.ReadKey();
var runner = session.GetRunner();
foreach (var o in this.outputs)
{
runner.Fetch(K.In(o).output);
}
foreach (var op in this.updates_op)
{
runner.AddTarget(op);
}
foreach (KeyValuePair <Tensor, Array> pair in feed_dict)
{
TensorFlowTensor t = K.In(pair.Key);
runner.AddInput(t.output, pair.Value);
}
var updated = runner.Run();
//Console.WriteLine();
//foreach (var v in updated)
//{
// object obj = v.GetValue();
// if (obj is float[,])
// Console.WriteLine((obj as float[,]).ToCSharp());
// else if (obj is float[])
// Console.WriteLine((obj as float[]).ToCSharp());
// else
// Console.WriteLine(obj);
//}
//Console.WriteLine();
//Console.WriteLine();
//Console.WriteLine("After:");
//PrintVariables(feed_dict, session);
return(updated.Get(0, this.outputs.Count).Select(t => K.Out(t)).ToList());
// Console.ReadKey();
}
