Add function for creating assymetric dummy tanimoto scores and predic…

…tions

tests/test_CalculateScoresBetweenAllIonmodes.py

@@ -3,31 +3,42 @@
 import numpy as np
 import pandas as pd
 
-from ms2deepscore.benchmarking.CalculateScoresBetweenAllIonmodes import CalculateScoresBetweenAllIonmodes, PredictionsAndTanimotoScores
-from ms2deepscore.benchmarking.plot_rmse_per_bin import plot_loss_per_bin, plot_loss_per_bin_multiple_benchmarks
+from ms2deepscore.benchmarking.CalculateScoresBetweenAllIonmodes import CalculateScoresBetweenAllIonmodes, \
+    PredictionsAndTanimotoScores
 from tests.test_user_worfklow import load_processed_spectrums, TEST_RESOURCES_PATH
 import random
 
 
-def dummy_tanimoto_scores(num_of_unique_inchikeys):
+def dummy_tanimoto_scores(num_of_unique_inchikeys, num_of_unique_inchikeys2):
     """Creates a dataframe with dummy tanimoto scores"""
     # Create list of dummy_inchikeys, with the same letter repeating
     dummy_inchikeys = [f"{14 * letter}" for letter in list(string.ascii_uppercase[:num_of_unique_inchikeys])]
+    dummy_inchikeys_2 = [f"{14 * letter}" for letter in list(string.ascii_uppercase[:num_of_unique_inchikeys2])]
+
     # Generate random values
-    random_values = np.random.rand(len(dummy_inchikeys), len(dummy_inchikeys))
-    # Make the matrix symmetric
-    symmetric_values = (random_values + random_values.T) / 2
+    random_values = np.random.rand(len(dummy_inchikeys), len(dummy_inchikeys_2))
     # Create the DataFrame with tanimoto_scores
-    tanimoto_scores = pd.DataFrame(symmetric_values, index=dummy_inchikeys, columns=dummy_inchikeys)
-    # Make sure all the diagonals are 1
-    np.fill_diagonal(tanimoto_scores.values, 1)
+    tanimoto_scores = pd.DataFrame(random_values, index=dummy_inchikeys, columns=dummy_inchikeys_2)
+
+    # Make the matrix symmetric
+    for inchikey in dummy_inchikeys:
+        for inchikey_2 in dummy_inchikeys_2:
+            if inchikey == inchikey_2:
+                tanimoto_scores[inchikey][inchikey_2] = 1.0
+            else:
+                try:
+                    tanimoto_scores[inchikey][inchikey_2] = tanimoto_scores[inchikey_2][inchikey]
+                except KeyError:
+                    continue
     return tanimoto_scores
 
 
-def create_dummy_predictions(tanimoto_scores: pd.DataFrame):
+def create_dummy_predictions_symmetric(tanimoto_scores: pd.DataFrame):
     """Creates a dataframe with predictions that has the same inchikeys as the given tanimoto scores,
     but for half of the tanimoto scores extra colums are added to mimick multiple spectra per inchikey"""
     predictions = tanimoto_scores.__deepcopy__()
+    if predictions.shape[0] != predictions.shape[1]:
+        raise ValueError("Predictions are expected to not be symmetric")
     inchikeys = list(predictions.index)
     random_half_of_inchikeys = random.sample(inchikeys, len(inchikeys)//2)
     for inchikey in random_half_of_inchikeys:
@@ -46,15 +57,43 @@ def create_dummy_predictions(tanimoto_scores: pd.DataFrame):
     return symmetric_predictions
 
 
-def create_dummy_predictions_and_tanimoto_scores(nr_of_unique_inchikeys):
-    tanimoto_scores = dummy_tanimoto_scores(nr_of_unique_inchikeys)
-    predictions = create_dummy_predictions(tanimoto_scores)
+def create_dummy_predictions_not_symmetric(tanimoto_scores: pd.DataFrame):
+    """Creates a dataframe with predictions that has the same inchikeys as the given tanimoto scores,
+    but for half of the tanimoto scores extra colums are added to mimick multiple spectra per inchikey"""
+    predictions = tanimoto_scores.__deepcopy__()
+    if predictions.shape[0] == predictions.shape[1]:
+        raise ValueError("Predictions are expected to not be symmetric")
+    inchikeys = list(predictions.index)
+    # Randomly duplicate half of the rows
+    for inchikey in random.sample(inchikeys, len(inchikeys)//2):
+        new_row = tanimoto_scores.loc[inchikey].copy()
+        new_row.name = inchikey  # Set the name for the new row
+        predictions = pd.concat([predictions, pd.DataFrame([new_row])])
+    # Randomly duplicate half of the columns
+    inchikeys_2 = list(predictions.columns)
+    for inchikey in random.sample(inchikeys_2, len(inchikeys_2)//2):
+        predictions["new_copy"] = predictions[inchikey]
+        predictions.rename(columns={'new_copy': inchikey}, inplace=True)
+    random_noise = np.random.normal(loc=0, scale=0.1, size=predictions.shape)
+    predictions = predictions + random_noise
+    # Set diagonal to nan
+    for i in range(min(tanimoto_scores.shape)):
+        predictions.iloc[i, i] = None
+    return predictions
+
+
+def create_dummy_predictions_and_tanimoto_scores(nr_of_unique_inchikeys, nr_of_unique_inchikeys_2):
+    tanimoto_scores = dummy_tanimoto_scores(nr_of_unique_inchikeys, nr_of_unique_inchikeys_2)
+    if nr_of_unique_inchikeys == nr_of_unique_inchikeys_2:
+        predictions = create_dummy_predictions_symmetric(tanimoto_scores)
+    else:
+        predictions = create_dummy_predictions_not_symmetric(tanimoto_scores)
     return PredictionsAndTanimotoScores(predictions, tanimoto_scores, True)
 
 
 def test_predictions_and_tanimoto_scores():
-    predictions_and_tanimoto_scores = create_dummy_predictions_and_tanimoto_scores(26)
-    average_loss = predictions_and_tanimoto_scores.get_average_loss_per_inchikey_pair()
+    predictions_and_tanimoto_scores = create_dummy_predictions_and_tanimoto_scores(26, 10)
+    average_loss = predictions_and_tanimoto_scores.get_average_MAE_per_inchikey_pair()
     average_MSE = predictions_and_tanimoto_scores.get_average_MSE_per_inchikey_pair()
     average_RMSE = predictions_and_tanimoto_scores.get_average_RMSE_per_inchikey_pair()
     print(average_RMSE)