Model Uncertainty prediction¶

Note:

This notebook extends the "Custom DataLoader for Imbalanced dataset" notebook

• In this notebook we will use the higly imbalanced Protein Homology Dataset from KDD cup 2004

* The first element of each line is a BLOCK ID that denotes to which native sequence this example belongs. There is a unique BLOCK ID for each native sequence. BLOCK IDs are integers running from 1 to 303 (one for each native sequence, i.e. for each query). BLOCK IDs were assigned before the blocks were split into the train and test sets, so they do not run consecutively in either file.
* The second element of each line is an EXAMPLE ID that uniquely describes the example. You will need this EXAMPLE ID and the BLOCK ID when you submit results.
* The third element is the class of the example. Proteins that are homologous to the native sequence are denoted by 1, non-homologous proteins (i.e. decoys) by 0. Test examples have a "?" in this position.
* All following elements are feature values. There are 74 feature values in each line. The features describe the match (e.g. the score of a sequence alignment) between the native protein sequence and the sequence that is tested for homology.

Initial imports¶

In [1]:

import pandas as pd
import numpy as np

from pytorch_widedeep import Trainer
from pytorch_widedeep.preprocessing import TabPreprocessor
from pytorch_widedeep.models import TabMlp, WideDeep
from pytorch_widedeep.dataloaders import DataLoaderImbalanced
from pytorch_widedeep.metrics import Accuracy, Recall, Precision, F1Score
from pytorch_widedeep.initializers import XavierNormal
from pytorch_widedeep.datasets import load_bio_kdd04

from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report

import time
import datetime

import warnings

warnings.filterwarnings("ignore", category=DeprecationWarning)

import pandas as pd import numpy as np from pytorch_widedeep import Trainer from pytorch_widedeep.preprocessing import TabPreprocessor from pytorch_widedeep.models import TabMlp, WideDeep from pytorch_widedeep.dataloaders import DataLoaderImbalanced from pytorch_widedeep.metrics import Accuracy, Recall, Precision, F1Score from pytorch_widedeep.initializers import XavierNormal from pytorch_widedeep.datasets import load_bio_kdd04 from sklearn.model_selection import train_test_split from sklearn.metrics import classification_report import time import datetime import warnings warnings.filterwarnings("ignore", category=DeprecationWarning)

/Users/javierrodriguezzaurin/.pyenv/versions/3.10.15/envs/widedeep310/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
  from .autonotebook import tqdm as notebook_tqdm

In [2]:

df = load_bio_kdd04(as_frame=True)
df.head()

df = load_bio_kdd04(as_frame=True) df.head()

Out[2]:

	EXAMPLE_ID	BLOCK_ID	target	4	5	6	7	8	9	10	...	68	69	70	71	72	73	74	75	76	77
0	279	261532	0	52.0	32.69	0.30	2.5	20.0	1256.8	-0.89	...	-8.0	1595.1	-1.64	2.83	-2.0	-50.0	445.2	-0.35	0.26	0.76
1	279	261533	0	58.0	33.33	0.00	16.5	9.5	608.1	0.50	...	-6.0	762.9	0.29	0.82	-3.0	-35.0	140.3	1.16	0.39	0.73
2	279	261534	0	77.0	27.27	-0.91	6.0	58.5	1623.6	-1.40	...	7.0	1491.8	0.32	-1.29	0.0	-34.0	658.2	-0.76	0.26	0.24
3	279	261535	0	41.0	27.91	-0.35	3.0	46.0	1921.6	-1.36	...	6.0	2047.7	-0.98	1.53	0.0	-49.0	554.2	-0.83	0.39	0.73
4	279	261536	0	50.0	28.00	-1.32	-9.0	12.0	464.8	0.88	...	-14.0	479.5	0.68	-0.59	2.0	-36.0	-6.9	2.02	0.14	-0.23

5 rows × 77 columns

In [3]:

# drop columns we won't need in this example
df.drop(columns=["EXAMPLE_ID", "BLOCK_ID"], inplace=True)

# drop columns we won't need in this example df.drop(columns=["EXAMPLE_ID", "BLOCK_ID"], inplace=True)

In [4]:

df_train, df_valid = train_test_split(
    df, test_size=0.2, stratify=df["target"], random_state=1
)
df_valid, df_test = train_test_split(
    df_valid, test_size=0.5, stratify=df_valid["target"], random_state=1
)

df_train, df_valid = train_test_split( df, test_size=0.2, stratify=df["target"], random_state=1 ) df_valid, df_test = train_test_split( df_valid, test_size=0.5, stratify=df_valid["target"], random_state=1 )

Preparing the data¶

In [5]:

continuous_cols = df.drop(columns=["target"]).columns.values.tolist()

continuous_cols = df.drop(columns=["target"]).columns.values.tolist()

In [6]:

# deeptabular
tab_preprocessor = TabPreprocessor(continuous_cols=continuous_cols, scale=True)
X_tab_train = tab_preprocessor.fit_transform(df_train)
X_tab_valid = tab_preprocessor.transform(df_valid)
X_tab_test = tab_preprocessor.transform(df_test)

# target
y_train = df_train["target"].values
y_valid = df_valid["target"].values
y_test = df_test["target"].values

# deeptabular tab_preprocessor = TabPreprocessor(continuous_cols=continuous_cols, scale=True) X_tab_train = tab_preprocessor.fit_transform(df_train) X_tab_valid = tab_preprocessor.transform(df_valid) X_tab_test = tab_preprocessor.transform(df_test) # target y_train = df_train["target"].values y_valid = df_valid["target"].values y_test = df_test["target"].values

Define the model¶

In [7]:

deeptabular = TabMlp(
    column_idx=tab_preprocessor.column_idx,
    continuous_cols=tab_preprocessor.continuous_cols,
    mlp_hidden_dims=[64, 32],
)
model = WideDeep(deeptabular=deeptabular, pred_dim=1)
model

deeptabular = TabMlp( column_idx=tab_preprocessor.column_idx, continuous_cols=tab_preprocessor.continuous_cols, mlp_hidden_dims=[64, 32], ) model = WideDeep(deeptabular=deeptabular, pred_dim=1) model

Out[7]:

WideDeep(
  (deeptabular): Sequential(
    (0): TabMlp(
      (cont_norm): Identity()
      (encoder): MLP(
        (mlp): Sequential(
          (dense_layer_0): Sequential(
            (0): Linear(in_features=74, out_features=64, bias=True)
            (1): ReLU(inplace=True)
            (2): Dropout(p=0.1, inplace=False)
          )
          (dense_layer_1): Sequential(
            (0): Linear(in_features=64, out_features=32, bias=True)
            (1): ReLU(inplace=True)
            (2): Dropout(p=0.1, inplace=False)
          )
        )
      )
    )
    (1): Linear(in_features=32, out_features=1, bias=True)
  )
)

In [8]:

trainer = Trainer(
    model,
    objective="binary",
    metrics=[Accuracy(), Precision(), F1Score(), Recall()],
    verbose=1,
)

trainer = Trainer( model, objective="binary", metrics=[Accuracy(), Precision(), F1Score(), Recall()], verbose=1, )

In [9]:

start = time.time()
trainer.fit(
    X_train={"X_tab": X_tab_train, "target": y_train},
    X_val={"X_tab": X_tab_valid, "target": y_valid},
    n_epochs=3,
    batch_size=32,
)

start = time.time() trainer.fit( X_train={"X_tab": X_tab_train, "target": y_train}, X_val={"X_tab": X_tab_valid, "target": y_valid}, n_epochs=3, batch_size=32, )

epoch 1: 100%|██████████| 3644/3644 [00:20<00:00, 175.39it/s, loss=0.0222, metrics={'acc': 0.9945, 'prec': 0.7565, 'f1': 0.6419, 'rec': 0.5574}]
valid: 100%|██████████| 456/456 [00:01<00:00, 252.36it/s, loss=0.0125, metrics={'acc': 0.9969, 'prec': 0.92, 'f1': 0.8035, 'rec': 0.7132}]  
epoch 2: 100%|██████████| 3644/3644 [00:20<00:00, 177.43it/s, loss=0.0119, metrics={'acc': 0.9968, 'prec': 0.9209, 'f1': 0.793, 'rec': 0.6962}] 
valid: 100%|██████████| 456/456 [00:01<00:00, 255.61it/s, loss=0.0121, metrics={'acc': 0.997, 'prec': 0.8972, 'f1': 0.8136, 'rec': 0.7442}] 
epoch 3: 100%|██████████| 3644/3644 [00:20<00:00, 176.07it/s, loss=0.0103, metrics={'acc': 0.9973, 'prec': 0.9312, 'f1': 0.8351, 'rec': 0.757}] 
valid: 100%|██████████| 456/456 [00:01<00:00, 259.70it/s, loss=0.0119, metrics={'acc': 0.997, 'prec': 0.8909, 'f1': 0.8201, 'rec': 0.7597}] 

In [10]:

pd.DataFrame(trainer.history)

pd.DataFrame(trainer.history)

Out[10]:

	train_loss	train_acc	train_prec	train_f1	train_rec	val_loss	val_acc	val_prec	val_f1	val_rec
0	0.022229	0.994468	0.756545	0.641866	0.557377	0.012473	0.996913	0.920000	0.803493	0.713178
1	0.011912	0.996767	0.920918	0.792971	0.696239	0.012088	0.996981	0.897196	0.813559	0.744186
2	0.010341	0.997341	0.931198	0.835106	0.756991	0.011884	0.997050	0.890909	0.820084	0.759690

"Normal" prediction¶

In [11]:

df_pred = trainer.predict(X_tab=X_tab_test)
print(classification_report(df_test["target"].to_list(), df_pred))
print("Actual predicted values:\n{}".format(np.unique(df_pred, return_counts=True)))

df_pred = trainer.predict(X_tab=X_tab_test) print(classification_report(df_test["target"].to_list(), df_pred)) print("Actual predicted values:\n{}".format(np.unique(df_pred, return_counts=True)))

predict: 100%|██████████| 456/456 [00:00<00:00, 689.36it/s]

              precision    recall  f1-score   support

           0       1.00      1.00      1.00     14446
           1       0.91      0.78      0.84       130

    accuracy                           1.00     14576
   macro avg       0.95      0.89      0.92     14576
weighted avg       1.00      1.00      1.00     14576

Actual predicted values:
(array([0, 1]), array([14465,   111]))

Prediction using uncertainty¶

In [12]:

df_pred_unc = trainer.predict_uncertainty(X_tab=X_tab_test, uncertainty_granularity=10)
print(classification_report(df_test["target"].to_list(), df_pred))
print(
    "Actual predicted values:\n{}".format(
        np.unique(df_pred_unc[:, -1], return_counts=True)
    )
)

df_pred_unc = trainer.predict_uncertainty(X_tab=X_tab_test, uncertainty_granularity=10) print(classification_report(df_test["target"].to_list(), df_pred)) print( "Actual predicted values:\n{}".format( np.unique(df_pred_unc[:, -1], return_counts=True) ) )

predict_UncertaintyIter: 100%|██████████| 10/10 [00:05<00:00,  1.86it/s]

              precision    recall  f1-score   support

           0       1.00      1.00      1.00     14446
           1       0.91      0.78      0.84       130

    accuracy                           1.00     14576
   macro avg       0.95      0.89      0.92     14576
weighted avg       1.00      1.00      1.00     14576

Actual predicted values:
(array([0.]), array([14576]))

In [13]:

df_pred_unc

df_pred_unc

Out[13]:

array([[9.99999821e-01, 1.77245539e-07, 0.00000000e+00],
       [1.00000000e+00, 8.29310925e-11, 0.00000000e+00],
       [9.99995947e-01, 4.06420531e-06, 0.00000000e+00],
       ...,
       [9.99999940e-01, 3.85314713e-08, 0.00000000e+00],
       [1.00000000e+00, 2.98146707e-09, 0.00000000e+00],
       [1.00000000e+00, 1.21332046e-12, 0.00000000e+00]])