diff --git a/BM_25_pyterrier.py b/BM_25_pyterrier.py
index 242fcd2..d8e4e89 100644
--- a/BM_25_pyterrier.py
+++ b/BM_25_pyterrier.py
@@ -4,32 +4,34 @@
 # for the next parts of the project
 
 import pyterrier as pt
-import pandas as pd
 import gzip
 import shutil
 import os as os
+from pyterrier.measures import *
+import pandas as pd
 
 # Init pyterrier
 pt.init()
 
 # Get MS Marco passages used in TREC-2019
 dataset = pt.get_dataset("trec-deep-learning-passages")
-
+print(dataset)
 # Get corpus
 pathCorpus = dataset.get_corpus()
+print(pathCorpus)
 
 # Get the index stemmed (Porter stemmer)
 path = dataset.get_index("terrier_stemmed")
 index = pt.IndexFactory.of(path)
 
 # Get the queries
-queries = dataset.get_topics("test-2019")
+queries = dataset.get_topics("test-2020")
 print("query examples")
 print(queries)
 print()
 
 # Get the qrels
-qrels = dataset.get_qrels("test-2019")
+qrels = dataset.get_qrels("test-2020")
 print("qrel examples:")
 print(qrels)
 print()
@@ -38,23 +40,25 @@
 bm25 = pt.BatchRetrieve(index, wmodel="BM25")
 
 # Run BM-25 on the whole test dataset
-pt.Experiment(
+results = pt.Experiment(
     [bm25],
     queries,
     qrels,
-    eval_metrics=["map", "recip_rank", "ndcg"],
+    eval_metrics=["map", "recip_rank", "ndcg", "recall"],
     save_dir="./",
     save_mode="overwrite",
     dataframe=True,
 )
+print(results)
 
 # Run BM-25 on a subset of queries
-queries_uni = queries.loc[queries["qid"] == str(156493)]
+queries_uni = queries.loc[queries["qid"] == str(1037496)]
+print(queries_uni)
 pt.Experiment(
     [bm25],
     queries_uni,
     qrels,
-    eval_metrics=["map", "recip_rank", "ndcg"],
+    eval_metrics=["map", "recip_rank", "ndcg", "recall"],
     perquery=True,
     dataframe=True,
 )
@@ -79,3 +83,5 @@
     with open("retrieved.txt", "wb") as f_out:
         shutil.copyfileobj(f_in, f_out)
 os.remove("BR(BM25).res.gz")
+
+print("The ranking is generated !")
diff --git a/evaluate.py b/evaluate.py
deleted file mode 100644
index 58cbbf1..0000000
--- a/evaluate.py
+++ /dev/null
@@ -1,48 +0,0 @@
-import sys
-from trectools import TrecQrel, TrecRun, TrecEval
-
-arguments = sys.argv
-
-print("format: evaluate.py retrieved.txt qrels.txt")
-print("Note: all files should be on the same local path")
-
-retrievedPath = "retrieved.txt"
-qrelsPath = "qrels.txt"
-if len(arguments) == 1:
-    print(
-        "Assuming the files are retrieved.txt and qrels.txt if no parameters are used"
-    )
-elif len(arguments) == 3:
-    retrievedPath = arguments[1]
-    qrelsPath = arguments[2]
-else:
-    print("Print wrong numbers of parameters !")
-    exit()
-
-print()
-
-
-def evaluation():
-    r1 = TrecRun(retrievedPath)
-
-    qrelstrec = TrecQrel(qrelsPath)
-
-    te = TrecEval(r1, qrelstrec)
-
-    p100 = te.get_precision(depth=100)
-    rr = te.get_reciprocal_rank(depth=100)
-    ndcg = te.get_ndcg(depth=100)
-    map_ = te.get_map(depth=100)
-    print("relevent docs: ", te.get_relevant_documents())
-    print("retrieved docs: ", te.get_retrieved_documents())
-    print("retrieved relevant docs: ", te.get_relevant_retrieved_documents())
-    print()
-    print("P@100 : \t", p100)
-    print("MRR@100 : \t", rr)
-    print("NDCG@100 : \t", ndcg)
-    print("MAP@100 : \t", map_)
-    print()
-
-
-if __name__ == "__main__":
-    evaluation()
diff --git a/reranking.py b/reranking.py
new file mode 100644
index 0000000..09492ee
--- /dev/null
+++ b/reranking.py
@@ -0,0 +1,92 @@
+from sentence_transformers import CrossEncoder
+import pyterrier as pt
+import pandas as pd
+
+PATH_TO_TOP_1000 = "retrieved.txt"
+OUTPUT_PATH = "x.txt"
+
+# Init
+pd.set_option("display.max_rows", None)
+pd.set_option("display.max_colwidth", 150)
+pt.init()
+
+dataset = pt.get_dataset("trec-deep-learning-passages")
+
+# Get the previously retrieved top 1000 (by a baseline method)
+retrieved = pd.read_csv(PATH_TO_TOP_1000, sep=" ")
+retrieved.columns = ["qid", "Q0", "docID", "rank", "score", "system"]
+print(retrieved.dtypes)
+print(retrieved.head(n=5))
+print()
+
+# Get the queries
+queries = dataset.get_topics("test-2020")
+queries = queries.astype({"qid": "int64", "query": "string"})
+print(queries.dtypes)
+print("query examples")
+print(queries.head(n=5))
+print()
+
+
+# Get the text corpus
+pathCorpus = dataset.get_corpus()
+print(pathCorpus[0])
+print("Load CSV...")
+corpus = pd.read_csv(pathCorpus[0], sep="\t")
+corpus.columns = ["docno", "text"]
+corpus = corpus.astype({"text": "string"})
+print("corpus examples:")
+print(corpus.head(n=5))
+print()
+
+model = CrossEncoder("cross-encoder/ms-marco-TinyBERT-L-2-v2", max_length=512)
+
+
+def getReranked(qid):
+    querytext = queries.loc[queries["qid"] == qid].iloc[0]["query"]
+    print("query text: ", querytext)
+    docIds = retrieved.loc[retrieved["qid"] == qid]["docID"]
+    print(docIds.head(n=5))
+    docs = corpus.loc[corpus["docno"].isin(docIds)]
+    print(docs.head(n=5))
+    print()
+
+    print("Predict...")
+    couples = [(querytext, docText) for docText in docs["text"]]
+    scores = model.predict(couples)
+    print(scores)
+
+    print("Sort...")
+    sorted_indices = [i[0] for i in sorted(enumerate(scores), key=lambda x: -x[1])]
+
+    top = docs.iloc[sorted_indices]
+    return top
+
+
+s = ""
+numberquery = 0
+for qid in retrieved["qid"].unique():
+    print(numberquery, " query processed ...")
+    numberquery += 1
+    top = getReranked(qid)
+    i = 0
+    for index, row in top.iterrows():
+        s += (
+            str(qid)
+            + " "
+            + "Q0"
+            + " "
+            + str(row["docno"])
+            + " "
+            + str(i)
+            + " "
+            + str(1 / (i + 1))
+            + " "
+            + "BERT"
+            + "\n"
+        )
+        i += 1
+
+
+with open(OUTPUT_PATH, "w+") as file:
+    file.write(s)