[Rspamd-Users] Is my neural module working after updating from 1.9.4 to 2.0

[Tamás Németh]

Dear all,

Yesterday I updated from Rspamd 1.9.4 to 2.0. My neural-specific
configuration is the following:

###########################
neural {
  train {
    ham_score = -3;
    max_usages = 20;
    spam_score = 10;
    learning_rate = 0.010000;
    max_iterations = 25;
    max_train = 1000;
  }
  timeout = 20;
}

symbols {
  }
    NEURAL_SPAM {
    weight = 2;
  }
}

logging {
  debug_modules [ "dkim_signing", "neural" ];
}
###########################

Before the update my neural module had learnt enough, so I often saw
emails tagged with the NEURAL_SPAM symbol. I understand that the
neural module had to start learning from zero since the update, but
deciding based on these log messages, I'm not sure neural learning
goes well:

###########################
2019-10-17 09:17:11 #4895(rspamd_proxy) <xd3un9>; neural;
neural.lua:1053: try_load_ann: process element default:default
2019-10-17 09:17:11 #4895(rspamd_proxy) <xd3un9>; neural;
neural.lua:858: no ANN for default:default in Redis key
rn_default_default_boeduyc4_0
2019-10-17 09:17:11 #4896(controller) <xd3un9>; neural;
neural.lua:1053: try_train_ann: process element default:default
2019-10-17 09:17:11 #4896(controller) <xd3un9>; neural;
neural.lua:966: check if ANN rn_default_default_boeduyc4_0 needs to be
trained
2019-10-17 09:17:11 #4896(controller) <xd3un9>; neural;
neural.lua:1053: try_load_ann: process element default:default
2019-10-17 09:17:11 #4896(controller) <xd3un9>; neural;
neural.lua:982: no need to learn ANN rn_default_default_boeduyc4_0 562
spam learn vectors (1000 required)
2019-10-17 09:17:11 #4896(controller) <xd3un9>; neural;
neural.lua:858: no ANN for default:default in Redis key
rn_default_default_boeduyc4_0
2019-10-17 09:17:11 #4898(normal) <xd3un9>; neural; neural.lua:1053:
try_load_ann: process element default:default
2019-10-17 09:17:11 #4898(normal) <xd3un9>; neural; neural.lua:912:
ann default:default is not changed, our version = 0, remote version =
0
2019-10-17 09:17:11 #4897(normal) <xd3un9>; neural; neural.lua:1053:
try_load_ann: process element default:default
2019-10-17 09:17:11 #4897(normal) <xd3un9>; neural; neural.lua:912:
ann default:default is not changed, our version = 0, remote version =
0
###########################

Doesn't mean that "no ANN for default:default in Redis key
rn_default_default_boeduyc4_0" message mean that something goes wrong
with neural learning, or can I just ignore it. Is it a warning
anyways?

Thank you in advance,

Németh, Tamás