Merge pull request #104 from code-for-venezuela/luis/dashboaord-app

Luis/dashboard app

Author: Edilmo

data/processed/huggingface/service_training_dataset.csv

@@ -7,7 +7,7 @@
     training_args={
         "per_device_train_batch_size" : 15,
         "per_device_eval_batch_size" : 15,      # change batch size if you're having memory problems
-        "num_train_epochs" : 3,
+        "num_train_epochs" : 50,
         "warmup_steps" : 1000,
         "save_steps" : 1000,
         "save_total_limit" : 1,
@@ -17,6 +17,7 @@
         "eval_accumulation_steps" : 1,
         "learning_rate" : 5e-07,
         "adafactor" : True,
+        "metric_for_best_model" : "accuracy"
     },
     columns=['title'],
     train_dataset_name="service_training_dataset.csv",  # note that this is the same experiment as relevance classification but using different parameters

experiments_samples/service_classification_experiment.py

@@ -27,5 +27,5 @@
 
 )
 
-exp = ClassifierExperiment.from_branch_and_experiment("samples", "classifier")
+exp = ClassifierExperiment.from_branch_and_experiment("samples", "relevance_classifier")
 exp.run_experiment(args)