float getScoreDeep(const vector<int> &ts) {
        float output = 0.0;
        for (int i=0; i < hidden; i++) {
            float score = 0.0;
            for (int j=0; j < 14; j++) {
                score += hiddenWeights[(i*14+j)*units+ts[j]];
            }
            score = relu(score);
            output += outputWeights[i] * score;
        }

        return tanh(output);
    }

=========================

this one is 6 times faster o.O

    float getScoreDeep(const vector<int> &ts) {
        float output = 0.0;
        int k = 26 * ts[0] + ts[1];
        int k1 = 64 * ts[2] + ts[3];
        int k2 = 64 * ts[4] + ts[5];
        int k3 = 64 * ts[6] + ts[7];
        int k4 = 64 * ts[8] + ts[9];
        int k5 = 64 * ts[10] + ts[11];
        int k6 = 64 * ts[12] + ts[13];
        for (int i=0; i < hidden; i++) {
            float score = cache[k * hidden + i] + cache1[k1 * hidden + i] + cache2[k2 * hidden + i] + cache3[k3 * hidden + i] + cache4[k4 * hidden + i] + cache5[k5 * hidden + i] + cache6[k6 * hidden + i];
            score = relu(score);
            output += outputWeights[i] * score;
        }

        return tanh(output);
    }