DL4J Examples
based on version 0.0.3.3.*
Repository of Deeplearning4J neural net examples:
- Convolutional Neural Nets
- Deep-belief Neural Nets
- Glove Example
- Restricted Boltzmann Machines
- Recurrent Neural Nets
- Recursive Neural Nets
- TSNE
- Word2Vec
Development
We are still developing and tuning these examples. If you notice issues, please log them, and if you want to contribute, submit a pull request. Input is welcome here.
Check the POM to confirm where these examples are running from. If it has SNAPSHOT in the dl4j and nd4j then git clone those repositories and build locally. Otherwise use Maven Repo. We are working to get this stabilized as quickly as possible.
Documentation
For more information, check out deeplearning4j.org and its JavaDoc.
Performance
Note: We are aware most examples need to be fixed. Code base is going through some adaptations and this should settle soon
| Model Name | Accuracy | F1 | Status | Training | Timing |
|---|---|---|---|---|---|
| CNNIris | 0.60 | 0.71 | Fix | full | |
| CNNMnist | 0.41 | 0.26 | Fix | full | |
| CNNMnist2 | 0.20 | 0.09 | Fix | batch | |
| DBNCreateData | 0.50 | 0.66 | Fix | batch | |
| DBNFullMnist | 0.39 | 0.20 | Fix | batch | 63m7.25s |
| DBNIris | 0.83 | 0.88 | Fix | full | 0m3.78s |
| DBNLFW | Fix | batch | |||
| DBNMnistSingleLayer | 0.35 | 0.18 | Fix | full | 0m0.08s |
| DBNSmallMnist | 0.54 | 0.29 | Fix | full | 0m0.09s |
| GloveRawSentence | NA | Fix | batch | 0m0.73s | |
| MLPBackpropIris | 0.55 | 0.70 | Tune | batch | 0m0.12s |
| RBMCreateData | NA | Fix | full | 0m0.09s | |
| RBMIris | NA | Fix | full | 0m6.12s | |
| RecurrentLSTMMnist | NA | Fix | batch | ||
| RecursiveAutoEncoder | NA | Fix | batch | ||
| RNTNTweets | 0.33 | Fix | batch | ||
| RNTNTweets2 | 0.33 | Fix | batch | ||
| TSNEBarnesHut | NA | Fix | NA | ||
| TSNEStandard | NA | Fix | NA | ||
| Word2VecRawText | NA | Fix | batch | 0m1.36s |
- using bash real for timing and running on single CPU with 2.3GHz processor
- Some networks need adjustments for seed to work (e.g. RNTN)