ag-news-lstm.json

{
    "task": "classify",
    "basedir": "ag-news",
    "batchsz": 100,
    "preproc": {
	"mxlen": 100,
	"clean": true
    },
    "features": [
	{
	    "name": "word",
	    "vectorizer": { "type": "token1d" },
	    "embeddings": { "label": "glove-840B" }
	},
	{
	    "name": "word2",
	    "vectorizer": { "type": "token1d" },
	    "embeddings": { "label": "w2v-gn" }
	}
    ],
    "backend": "tensorflow",
    "dataset": "AG",
    "loader": {
	"reader_type": "default"
    },
    "unif": 0.25,
    "model": {
	"model_type": "lstm",
	"rnnsz": 200,
	"dropout": 0.5,
	"finetune": true
    },
    "train": {
	"epochs": 10,
	"optim": "adadelta",
	"eta": 1.0,
	"early_stopping_metric": "acc"
    }
}