-
Notifications
You must be signed in to change notification settings - Fork 5
Home
DreamingRaven edited this page Aug 30, 2018
·
7 revisions
Nemesyst is a highly configurable, generalised, deep learning framework for sequence based models. The deep learning building blocks availiable at this time are:
- Recurrent Neural Networks (RNNs)
- Long Short Term Memory Networks (LSTMs)
- Generative Adversarial Networks (GANs) - experimental
Distributed deep learning across machines/ servers, and quick single machine deep learning prototyping.
- To make applying deep learning to real world scenarios quick, and efficient
- While still allowing as much control to the inner mechanisms as possible
- To do things the Arch way
| option | alternate | default | isFlag | class | description |
|---|---|---|---|---|---|
| --coll | -C | "testColl" | 0 | mongo | sets collection to operate on |
| --cleaner | -c | **/examples/cleaner.py | 0 | import | specifies path to executable cleaner file |
| --dir | -D | ~/db | 0 | mongo | specifies path to mongoDb files |
| --newData | -d | None | 0 | import | specifies path to .csv data folder |
| --ip | -I (eye) | 127.0.0.1 | 0 | mongo | specifies ip of database |
| --toInitDb | -i | False | 1 | mongo | flags new user auth to create |
| --toLogin | -l (ell) | False | 1 | mongo | flags to log user into db for them |
| --name | -N | "RecSyst" | 0 | mongo | specifies the name of the mongoDb DB |
| --port | -P | 27017 | 0 | mongo | specifies the mongoDb port |
| --pass | -p | iamgroot | 0 | auth | specifies the password for mongoDb auth |
| --toStartDb | -S | False | 1 | mongo | flags to start mongoDb (with auth) |
| --toStopDb | -s | False | 1 | mongo | flags to stop db in -D path |
| --toTrain | -T | False | 1 | ann | flags to train |
| --toTest | -t | False | 1 | ann | flags to test |
| --url | -U | mongodb://localhost:27017/ | 0 | mongo | specifies mongoDb url |
| --user | -u | groot | 0 | auth | specifies the mongoDb usernam for auth |
| --loglevel | -v | 2 | 0 | debug | specifies the min loglevel to log |
| --batchSize | 1 | 0 | mongo ann | specifies the size of batches to use | |
| --target | target | 0 | mongo ann | specifies the name of target feature | |
| --type | lstm | 0 | ann | specifies the type of neural network to create | |
| --timeSteps | 25 | 0 | ann | specifies if sequential type, num cells of rnn | |
| --testSize | 0.2 | 0 | validation | specifies the % size of test set | |
| --activation | tanh | 0 | ann | specifies keras activation alg | |
| --dimensionality | 0 | ann | specifies num of features in data during learning and predicting | ||
| --layers | 1 | 0 | ann | specifies num of layers in things like lstm | |
| --lossMetic | mae | 0 | ann | specifies keras loss metric | |
| --optimizer | sgd | 0 | ann | specifies keras optimiser | |
| --randomSeed | 42 | 0 | ann | specifies random seed (unused) | |
| --epochs | 1 | 0 | ann | specifies num of keras epochs | |
| --suffix | .data | 0 | rrs | specifies extension of all temporary data files | |
| --chunkSize | 10000000 | 0 | rrs | specifies the maximum number of rows to be processed in imported csv file at a time | |
| --toJustImport | False | 1 | rrs | flags using residual temporary files without cleaning to import to db | |
| --pipeline | **/config/pipeline.json | 0 | config | specifies file path to pipeline.json file | |
| --config | **/config/config.ini | 0 | config | specifies file path to config.ini file | |
| --mongoCursorTimeout | 600 | 0 | mongo | specifies the time in milliseconds to allow a cursor to remain inacive before it is deleted |