diff --git a/admin/experiment.py b/admin/experiment.py
new file mode 100644
index 0000000..380a8bf
--- /dev/null
+++ b/admin/experiment.py
@@ -0,0 +1,50 @@
+import pandas as pd
+from functools import reduce
+from NERDA.datasets import get_conll_data
+from NERDA.models import NERDA
+
+def train_n(n = 5):
+    
+    print(f'Training with {n} observations')
+    model = NERDA(transformer = 'google/electra-small-discriminator',
+                  tag_scheme = [
+                      'B-PER',
+                      'I-PER', 
+                      'B-ORG', 
+                      'I-ORG', 
+                      'B-LOC', 
+                      'I-LOC', 
+                      'B-MISC', 
+                      'I-MISC'
+                      ],
+                  tag_outside = 'O',
+                  dataset_training = get_conll_data('train', n),
+                  dataset_validation = get_conll_data('valid'),
+                  max_len = 128,
+                  dropout = 0.1,
+                  hyperparameters = {'epochs' : 4,
+                                      'warmup_steps' : 250,
+                                      'train_batch_size': 13,
+                                      'learning_rate': 8e-05},
+                  tokenizer_parameters = {'do_lower_case' : True})
+
+    model.train()
+
+    performance = model.evaluate_performance(get_conll_data('test'))
+
+    performance = performance.append({'Level': 'N_SENTS', 'F1-Score' : n}, ignore_index=True)
+
+    performance = performance.rename(columns = {"F1-Score" : f'N{n}'})
+
+    print(performance)
+
+    return performance
+
+#training_sizes = [5, 10]
+#out = [train_n(n) for n in training_sizes]
+
+out = [train_n(n) for n in range(1039, len(get_conll_data('train')['sentences']) + 1000, 1000)]
+
+out = reduce(lambda x, y: pd.merge(x, y), out)
+
+out.to_csv("performance_versus_training_size.csv", index = False, sep = ";")