Backpropagation Neural Networks producing negative output

[Former Member]

Hi Everyone,

I am trying to model the commodity prices using neural networks. I have 20 independent variables and one target.

When I use Statistica package to run experiments, I was able to achieve predictions up to 97% accuracy. However, my goal is to embedded the network into a real time predictive analysis. Hence I am trying to do the same with SAP HANA.

However, with the predictive analytics library in SAP HANA I am getting negative decimal values like (-0.3568941235648891) but the value I am trying to predict ranges from 1500 to 3800.

Below are the settings for building my model

CREATE LOCAL TEMPORARY COLUMN TABLE "#PARAMS" LIKE "T_PARAMS";

INSERT INTO "#PARAMS" VALUES ('HIDDEN_LAYER_ACTIVE_FUNC', 3, null, null);

INSERT INTO "#PARAMS" VALUES ('OUTPUT_LAYER_ACTIVE_FUNC', 3, null, null);

INSERT INTO "#PARAMS" VALUES ('LEARNING_RATE', null, 0.1, null);

INSERT INTO "#PARAMS" VALUES ('MOMENTUM_FACTOR', null, 0.1, null);

INSERT INTO "#PARAMS" VALUES ('HIDDEN_LAYER_SIZE', null, null, '1');

INSERT INTO "#PARAMS" VALUES ('MAX_ITERATION', 500, null, null);

INSERT INTO "#PARAMS" VALUES ('FUNCTIONALITY', 1, null, null); -- 0:Classification; 1:Regression

INSERT INTO "#PARAMS" VALUES ('TARGET_COLUMN_NUM', 1, null, null);

INSERT INTO "#PARAMS" VALUES ('TRAINING_STYLE', 0, null, null); -- 0:Batch; 1:Stochastic

INSERT INTO "#PARAMS" VALUES ('NORMALIZATION', 1, null, null); -- 0:Normal; 1:Z-transform; 2:Scalar

INSERT INTO "#PARAMS" VALUES ('WEIGHT_INIT', 0, null, null); -- 0:all zeros; 1: normal; 2: uniform

--INSERT INTO "#PARAMS" VALUES ('CATEGORY_COL', 0, null, null);

Please is there something I am doing wrong here.

Thank you.

[achab]

Hi Kayode,

Are you using our SAP Predictive Analytics software or are you doing the coding directly in SAP HANA? Just to make sure your post is on the right forum for people to look at it.

Thanks & regards

Antoine

[Former Member]

Hi Antoine,

I am using SAP HANA and writing codes to generate the model.

is that out of scope here?

[achab]

Hi Kayode,

I was thinking of this community here: as another place where you might find help.

Thanks & regards

Antoine

[Former Member]

Thanks Antoine.