I wrote a blog post to explain the detail.
| Parameter | From Scratch | Tensorflow (CPU) |
|---|---|---|
| batch size | 128 | 128 |
| embedding size | 30 | 30 |
| num sampled | 10 | 10 |
| num steps | 70001 | 1500001 |
| learning rate | 0.025 | 1 |
# from scratch # tensorflow
# spend: 53.94 min # spend: 38.09 min
雲 1.0 雲 1.0
嵐 0.818097894953 嵐 0.731922132211
緲 0.807170161919 霞 0.710187307407
烽 0.806751349354 烟 0.693668808384
烟 0.791932317029 雪 0.684637639979
靄 0.790464066718 虹 0.683235227787
----- -----
峰 1.0 峰 1.0
峯 0.96521154438 峯 0.942029995583
層 0.869375215503 嶽 0.73387296403
巒 0.847521841138 嵋 0.732944525
巖 0.842055300736 巒 0.716149847575
巔 0.834164942036 巔 0.714281751101
----- -----
風 1.0 風 1.0
飆 0.839413385589 吹 0.820511746298
涼 0.812897226871 飆 0.809179019451
凜 0.790959089145 逆 0.67986909613
颸 0.786966264664 颸 0.663089281948
暄 0.771490669881 涼 0.659044072466
----- -----
女 + 父 - 男 女 + 父 - 男
母 0.765840473955 母 0.735594336365
婦 0.758031202523 子 0.729155945201
子 0.724152991944 伴 0.696736003898
伴 0.707958812532 彿 0.645417693955
阿 0.702062120972 阿 0.629788529922