Open personal-coding opened 4 years ago
The config I am using is this
{
"agent_type": "NNAgent",
"input": {
"coin_number": 11,
"coins": [
"ETH",
"XRP",
"BCH",
"LTC",
"STR",
"XMR",
"DASH",
"ETC",
"XEM",
"NXT",
"EMC2"
],
"end_date": "2019/06/01",
"fake_ratio": 1,
"feature_number": 3,
"global_period": 1800,
"is_permed": false,
"market": "poloniex",
"norm_method": "absolute",
"online": false,
"portion_reversed": false,
"save_memory_mode": false,
"start_date": "2016/01/01",
"test_portion": 0.08,
"volume_average_days": 30,
"window_size": 30
},
"layers": [
{
"activation_function": "relu",
"filter_number": 2,
"filter_shape": [
1,
3
],
"padding": "valid",
"regularizer": null,
"strides": [
1,
1
],
"type": "ConvLayer",
"weight_decay": 0
},
{
"activation_function": "relu",
"filter_number": 10,
"regularizer": "L2",
"type": "EIIE_Dense",
"weight_decay": 5e-9
},
{
"regularizer": "L2",
"type": "EIIE_Output_WithW",
"weight_decay": 5e-8
}
],
"random_seed": 0,
"trading": {
"buffer_biased": 0.00005,
"learning_rate": 0.00028,
"rolling_training_steps": 85,
"trading_consumption": 0.0025
},
"training": {
"batch_size": 160,
"buffer_biased": 0.00005,
"decay_rate": 1,
"decay_steps": 50000,
"fast_train": true,
"learning_rate": 0.00028,
"loss_function": "loss_function6",
"snap_shot": false,
"steps": 80000,
"training_method": "GradientDescent"
}
}
And here are my results.
it seems to me that the agent is stuck on a local minimum or fail to initialize
Is the user guide plot still accurate? My tests are generally showing break-even return.