AutoML for Regression Model Screening

The following are the steps for using AutoML for a regression task:

Provide an input pandas dataframe consisting of two columns:
- SMILES strings
- target class values
If featurization = True, Molecules are represented as:
- Coulomb matrix
- RDKit Morgan (ECFP6 1024-bit) fingerprints
- MACCs keys
- RDKit hashed topological torsion
- RDKit molecular descriptors (all)
- Mordred molecular descriptors
If featurization=False, a feature set must be provided by the user.
Screens through various regressor models:
- Single-core (runs regardless of multi_core value):
  - Ridge
  - Lasso
  - SVR (if n_datapoints<500)
  - ElasticNet
  - DecisionTreeRegressor
- Multi-core(set multi_core=True):
  - MLPRegressor (if n_datapoints<=500)
  - GradientBoostingRegressor (if n_datapoints<=500)
  - RandomForestRegressor
  - ChemML_MLP (if n_datapoints>500)
  - (optional) XGBRegressor
  - (optional) LightGBM

Yields ‘n-best’ models, with optimized hyperparamters.

The output_dir argument specifies a folder that will be created wherever the code is run from, containing the following directory tree:

 <output_dir>/
 |-- output.log
 |-- scores.csv
 |-- best_model/
 |   |-- model.pkl # most non-MLP models
 |   |-- model_chemml_model.json # MLP case (plus possible extra model files)
 |   |-- scaler.pkl # present when scaler exists for chosen feature_key
 |   |-- feature_order.json
 |   |-- metadata.json
 |-- .model_screener_artifacts/
 |   |-- <run_key_1>/
 |        |-- model.pkl # or MLP artifact files
 |   |-- <run_key_2>/
...

Load your data

We are using ChemML’s in-built dataset to demonstrate the minimum requirements to run AutoML.

[1]:

import pandas as pd
from chemml.datasets import load_organic_density

[2]:

molecules, target, dragon_subset = load_organic_density()
df=pd.concat([molecules, target], axis=1)
df = df.sample(200, random_state=42)
df

[2]:

	smiles	density_Kg/m3
361	c1scc(n1)c1ncc(s1)c1c(ccc2c1cccc2)c1cncs1	1328.35
73	S1CC(SC(C1)C1CCCC1)c1cocc1c1ccccc1	1143.80
374	N1CNC(NC1)c1sccc1C1(CSCCS1)c1cccs1	1351.33
155	c1nc(c(s1)c1nsnc1)c1csc(n1)c1cccs1	1466.57
104	Oc1ccc2c(c1c1cocc1)c(ccc2)c1ccccn1	1207.67
...	...	...
248	Oc1ncc(c(c1)C)c1cnccn1	1195.16
145	Sc1cc2ccccc2c(c1)c1cnccc1c1cnccn1	1213.41
92	c1sc(c(c1)c1ncccc1c1nsnc1)c1ccc2c(c1)cccc2	1264.16
152	N1CNC(NC1)(N1CNC(NC1)c1cncs1)c1cccc2c1cccc2	1270.40
367	C1CSC(CS1)C1(CSC(CS1)c1ccc2c(c1)cccc2)c1cnccn1	1268.28

200 rows × 2 columns

Run autoML for a regression task

Since we have not provided any feature set, we can use AutoML’s in-built featurization to screen through feature representations and find the most suitable feature set for our data.

[3]:

from chemml.autoML import ModelScreener
MS = ModelScreener(df, target="density_Kg/m3", featurization=True, smiles="smiles",
                   screener_type="regressor", output_dir="automl_results")
scores = MS.screen_models(n_best=10, multi_core=False)



-------------------------ChemML AutoML Model Screening-------------------------

-------------------------Model screening started at Mon Aug 31 01:37:15 2026-------------------------



Converting SMILES to ChemML Molecule objects...

Converting SMILES to ChemML Molecule objects: 100%|██████████| 200/200 [00:06<00:00, 29.03it/s]


Generating Coulomb matrix representation...

featurizing molecules in batches of 12 ...
200/200 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 13s 66ms/step
Merging batch features ...    [DONE]

Generating RDKit fingerprint representations...

Generating Morgan Fingerprints: 100%|██████████| 200/200 [00:00<00:00, 8905.58molecule/s]
Generating MACCS Fingerprints: 100%|██████████| 200/200 [00:00<00:00, 1264.05molecule/s]
Generating Hashed Topological Torsion Fingerprints: 100%|██████████| 200/200 [00:00<00:00, 6718.20molecule/s]
Generating Atom Pair Fingerprints: 100%|██████████| 200/200 [00:00<00:00, 7265.51molecule/s]


Generating RDKit descriptor representations...

Step 1: Normalizing molecule input

Step 2: Calculating RDKit descriptors: 100%|██████████| 200/200 [00:02<00:00, 94.76it/s]


Generating Mordred descriptor representations...

Step 1: Normalizing molecule input

Normalizing Molecule input: 100%|██████████| 200/200 [00:00<00:00, 514007.84it/s]

Step 2/3: Optimizing geometry for 3D descriptor calculation

3D geometry already exists, loading geometry: 100%|██████████| 200/200 [00:00<00:00, 77129.53it/s]

Step 3/3: Calculating Mordred descriptors


100%|██████████| 200/200 [00:06<00:00, 32.11it/s]


Cleaning feature sets to remove constant, highly correlated, low-variance, and complex features...

Feature set 'CoulombMatrix' cleaned: 25 features retained.

Feature set 'morganfingerprints_radius3' cleaned: 252 features retained.

Feature set 'MACCS_radius3' cleaned: 48 features retained.

Feature set 'hashedtopologicaltorsion_radius3' cleaned: 76 features retained.

Feature set 'hashedatompair_radius3' cleaned: 421 features retained.

Feature set 'rdkit_descriptors' cleaned: 90 features retained.

Feature set 'mord_descriptors' cleaned: 530 features retained.


-------------------------Run parameters-------------------------
  Featurization: True
  Multi_core: False
  Screener_type: regressor
  Number of datapoints: 200
  Note: Dataset <= 500 samples; MLP is excluded from screening due to inefficiency.

split done!
Single-core complete


------------------------- Screening complete for feature set CoulombMatrix, time taken: 10.8 seconds -------------------------

split done!
Single-core complete


------------------------- Screening complete for feature set morganfingerprints_radius3, time taken: 10.383 seconds -------------------------

split done!
Single-core complete


------------------------- Screening complete for feature set MACCS_radius3, time taken: 11.583 seconds -------------------------

split done!
Single-core complete


------------------------- Screening complete for feature set hashedtopologicaltorsion_radius3, time taken: 15.719 seconds -------------------------

split done!
Single-core complete


------------------------- Screening complete for feature set hashedatompair_radius3, time taken: 12.056 seconds -------------------------

split done!
Single-core complete


------------------------- Screening complete for feature set rdkit_descriptors, time taken: 21.011 seconds -------------------------

split done!
Single-core complete


------------------------- Screening complete for feature set mord_descriptors, time taken: 13.296 seconds -------------------------

[4]:

scores

[4]:

	ME	MAE	MSE	RMSE	MSLE	RMSLE	MAPE	MaxAPE	RMSPE	MPE	MaxAE	deltaMaxE	r_squared	time(seconds)	Model	parameters	Feature	run_key
33	4.552818	11.553280	181.728198	13.480660	0.000114	0.010676	0.910524	2.579643	1.062078	0.353552	33.017888	45.542634	0.980328	4.930333	Lasso	{'alpha': 0.10000000000000002, 'copy_X': True,...	mord_descriptors	Lasso_mord_descriptors_3a7194cd47ba458d8ba38a4...
30	3.515362	10.915172	188.539081	13.730953	0.000124	0.011143	0.869834	2.193016	1.108684	0.290819	28.069291	47.432492	0.979591	1.207374	Ridge	{'alpha': 63.800000000000004, 'copy_X': True, ...	mord_descriptors	Ridge_mord_descriptors_c196269642bc48e1ac9f745...
34	3.687586	10.678849	191.563700	13.840654	0.000122	0.011051	0.843186	2.740330	1.099991	0.299250	35.074581	57.654403	0.979263	5.995100	ElasticNet	{'alpha': 0.06951927961775611, 'copy_X': True,...	mord_descriptors	ElasticNet_mord_descriptors_516936b2cb4d4804af...
31	4.483152	11.853263	232.867162	15.259986	0.000144	0.012013	0.928513	2.919669	1.198157	0.343363	37.370005	66.251821	0.974792	2.749953	SVR	{'C': 1.0, 'cache_size': 200, 'coef0': 0.0, 'd...	mord_descriptors	SVR_mord_descriptors_f44f5f430c6043969a4062edb...
27	4.182179	16.889614	476.895174	21.837930	0.000284	0.016856	1.318434	4.192188	1.674893	0.308835	61.499391	97.646100	0.948376	1.832462	Lasso	{'alpha': 0.10000000000000002, 'copy_X': True,...	rdkit_descriptors	Lasso_rdkit_descriptors_91aafe99bf1344fe8545c7...
26	4.139097	17.790071	496.914400	22.291577	0.000297	0.017234	1.389557	4.205474	1.713232	0.310069	61.694307	98.196240	0.946209	0.783303	Ridge	{'alpha': 5.0, 'copy_X': True, 'fit_intercept'...	rdkit_descriptors	Ridge_rdkit_descriptors_8e884cb01ad148408712ef...
28	4.401149	18.037694	503.715533	22.443608	0.000301	0.017348	1.408142	4.234329	1.724064	0.330047	62.117602	98.276555	0.945473	1.884814	ElasticNet	{'alpha': 0.10000000000000002, 'copy_X': True,...	rdkit_descriptors	ElasticNet_rdkit_descriptors_19bde069eb2649ecb...
29	5.004877	19.254586	623.825962	24.976508	0.000380	0.019482	1.513790	4.573626	1.934141	0.382821	67.095092	105.112449	0.932471	13.696179	SVR	{'C': 11.421052631578947, 'cache_size': 200, '...	rdkit_descriptors	SVR_rdkit_descriptors_b46b8047cc1a428298ed35da...
25	7.912750	27.688750	1137.561421	33.727754	0.000713	0.026710	2.159354	6.548075	2.645154	0.525923	78.700000	134.570000	0.876860	1.285904	DecisionTreeRegressor	{'ccp_alpha': 0.0, 'criterion': 'absolute_erro...	rdkit_descriptors	DecisionTreeRegressor_rdkit_descriptors_b102e2...
21	-4.632258	25.970876	1146.052758	33.853401	0.000693	0.026329	2.024358	7.226699	2.680475	-0.414917	87.839081	133.848639	0.875940	1.143952	Ridge	{'alpha': 78.5, 'copy_X': True, 'fit_intercept...	hashedatompair_radius3	Ridge_hashedatompair_radius3_7c27bb60e96049318...

Validating AutoML-generated model

Since ChemML v1.3.4, AutoML outputs the best model and required scaler and feature order as easy-to-import files. A quick demonstration of the model loading workflow is given below:

[5]:

import pickle
import json

with open('automl_results/best_model/metadata.json', 'r') as f:
    metadata = json.load(f)
with open('automl_results/best_model/explanation.json', 'r') as f:
    explanation = json.load(f)
with open('automl_results/best_model/feature_order.json', 'r') as f:
    feature_order = json.load(f)
with open("automl_results/best_model/model.pkl", "rb") as f:
    loaded_model = pickle.load(f)
with open("automl_results/best_model/scaler.pkl", "rb") as f:
    loaded_scaler = pickle.load(f)

Metadata isn’t necessary for model validation, but gives us a lot of context from the AutoML screening workflow.

[6]:

metadata

[6]:

{'artifact_schema_version': 1,
 'model_name': 'Lasso',
 'feature_key': 'mord_descriptors',
 'run_key': 'Lasso_mord_descriptors_3a7194cd47ba458d8ba38a4c42e3dc6d',
 'screener_type': 'regressor',
 'parameters': {'alpha': 0.10000000000000002,
  'copy_X': True,
  'fit_intercept': True,
  'max_iter': 1000,
  'positive': False,
  'precompute': False,
  'random_state': None,
  'selection': 'cyclic',
  'tol': 0.0001,
  'warm_start': False},
 'n_folds': 5,
 'n_datapoints': 200,
 'metrics': {'ME': 4.552817883339197,
  'MAE': 11.553280063774162,
  'MSE': 181.72819797227447,
  'RMSE': 13.480660146011934,
  'MSLE': 0.00011396673759956667,
  'RMSLE': 0.010675520483778142,
  'MAPE': 0.9105242223102075,
  'MaxAPE': 2.5796433993517534,
  'RMSPE': 1.0620775201892783,
  'MPE': 0.35355188263843224,
  'MaxAE': 33.01788772566283,
  'deltaMaxE': 45.5426336527853,
  'r_squared': 0.980328032332101,
  'time(seconds)': 4.930333375930786},
 'model_serializer': 'pickle',
 'model_artifact_dir': 'model_artifact',
 'scaler_file': 'scaler.pkl',
 'feature_order_file': 'feature_order.json',
 'timestamp': 'Mon Aug 31 01:39:25 2026'}

Explanation gives the user a list of feature importances alongside feature names; while not necessary for validation, this is useful to understand what features the model is actually using

[7]:

explanation

[7]:

{'model_type': 'Lasso',
 'explanation_type': 'linear.coef_',
 'importance_values': {'AATSC0Z': 28.012900357435722,
  'AATS1se': 27.583120713121403,
  'AMID_h': 12.055038066747914,
  'MIC2': 10.045620227882683,
  'GATS2i': -9.042124462668491,
  'GATS1p': -7.354977238746177,
  'SpDiam_A': 6.726688403957051,
  'C2SP3': -5.591252791033305,
  'Mor06p': 5.2500775423204304,
  'AATS8dv': 4.8109148833883,
  'ZMIC4': 4.683909744557295,
  'GATS7s': -4.2733394067530215,
  'NaaN': 4.0778101734185075,
  'NsCH3': -4.031087808617084,
  'ATSC1pe': -3.9878753533945037,
  'Mor27m': 3.9049517797044606,
  'ATSC2d': -3.6164709139902325,
  'MDEC-22': -3.3562645191398004,
  'MATS7s': -3.3299319725002325,
  'Mor31': -3.233335107924634,
  'VSA_EState1': 3.096026758544977,
  'AATSC3Z': -2.9299423458120306,
  'AATS7se': 2.8514396676457734,
  'GATS8Z': 2.7909016705914294,
  'ATSC1i': -2.6094166454339014,
  'Mor08': -2.3210227304435325,
  'Mor32m': 2.311053179367852,
  'AATS7dv': 2.2720719610502043,
  'GATS6c': -2.239558871212753,
  'GATS3Z': -2.182318800354107,
  'GeomRadius': 2.1622850010862686,
  'Mor06': -2.126688194238739,
  'GATS6i': 2.0877098445975775,
  'ATSC6c': -2.0810486982550613,
  'Mor18m': -1.9553538109138944,
  'ATS7Z': -1.9315114841892362,
  'MATS8s': -1.896337434988026,
  'MATS7se': -1.8934550073453553,
  'ATSC4se': 1.7650120189269423,
  'GATS7c': -1.641581287765814,
  'Mor26m': 1.577904395408391,
  'ATSC4i': -1.566327233570521,
  'GATS4dv': 1.522849838289012,
  'Mor08m': 1.498998724869166,
  'GGI6': -1.4887000990735948,
  'ATSC5Z': -1.4744774617897087,
  'ATSC0Z': 1.4563678200672814,
  'Mor30m': -1.4366792446800893,
  'BCUTd-1h': -1.4047179154889509,
  'GATS1v': -1.4010746192631953,
  'GATS4d': 1.3546510016250888,
  'AATS6p': 1.3402285499839746,
  'EState_VSA3': 1.331851877186613,
  'ETA_beta_ns_d': 1.2472424735125094,
  'EState_VSA10': 1.2460205319655364,
  'PEOE_VSA10': 1.2193539938482074,
  'ATSC7se': 1.2136992014233667,
  'Mor12m': 1.1958135876100358,
  'ATSC8pe': -1.1874359616012609,
  'GATS4p': 1.1809986175118783,
  'AATSC2Z': -1.1762628322771915,
  'EState_VSA9': 1.135827903654703,
  'GeomDiameter': 1.134749434867131,
  'Mor19': -1.0368299813951307,
  'Mor03': -0.9908395783872673,
  'ATSC8v': 0.9623213981341334,
  'Mor17': 0.9300665599352823,
  'AATS6se': 0.8776051158610542,
  'Mor16m': 0.8636911078931104,
  'BCUTdv-1l': 0.7824838660887756,
  'BCUTs-1l': -0.7790779225419351,
  'VSA_EState7': -0.7231425759826096,
  'Mor23v': 0.6864958939020279,
  'ATSC8s': -0.6492842071511602,
  'MATS1c': -0.6110958331929972,
  'RPCS': 0.6061807410396066,
  'Mor13': 0.5771621621329794,
  'Mor29': 0.5729484951040458,
  'BCUTv-1l': -0.5364912859183224,
  'Mor27': -0.5199838003581592,
  'Mor14m': -0.49885083042878037,
  'RNCS': -0.4709699612068487,
  'ATSC7i': 0.4564951804103089,
  'GATS5p': 0.45445658818812107,
  'ATSC3Z': -0.4541469044094732,
  'GATS3pe': -0.4534761037497193,
  'GATS3se': -0.45320873550405516,
  'Mor09': 0.44067677473814004,
  'BCUTdv-1h': -0.4331796206217456,
  'Mor13m': 0.4311583368360955,
  'MOMI-Z': -0.4168298077982454,
  'SMR_VSA9': -0.39609316613612233,
  'BCUTi-1h': 0.39595003084238045,
  'ZMIC2': 0.3559846428336694,
  'ATSC4s': 0.35136005701664563,
  'Mor21m': -0.34944270053999066,
  'MOMI-X': 0.34024511935172036,
  'Mor05': -0.33714117682027256,
  'ATSC3se': -0.33030676013484417,
  'Mor28': -0.30707702153951827,
  'n6aHRing': -0.26684187962901446,
  'ATSC1dv': -0.257855333466701,
  'SssssC': 0.2568690218303352,
  'Mor24': 0.2339989848426439,
  'ATSC4c': 0.20990132236157932,
  'SlogP_VSA11': 0.2050387554046178,
  'EState_VSA2': 0.17029503917217784,
  'ATSC4v': -0.16899175791863136,
  'MATS8c': 0.16177124282143437,
  'C3SP3': 0.13708082063127403,
  'AATS6d': 0.10563704026892254,
  'ATSC1s': -0.10188989724213939,
  'SM1_DzZ': 0.08479288306819252,
  'Mor05v': -0.06290731676508811,
  'Mor10': 0.05208194143342154,
  'ATSC5se': 0.04629656719959114,
  'AATS8se': 0.04371660784846001,
  'AATS6Z': 0.0433391371869164,
  'PEOE_VSA7': -0.023701661132782152,
  'Mor04': 0.00979816509030612,
  'ATSC4pe': 0.002006335217403847,
  'ZMIC3': 0.000210759590768856,
  'ABC': 0.0,
  'ABCGG': 0.0,
  'nBase': 0.0,
  'nAromAtom': 0.0,
  'nAtom': -0.0,
  'nHetero': 0.0,
  'nH': -0.0,
  'nC': -0.0,
  'nN': 0.0,
  'nO': -0.0,
  'nS': 0.0,
  'ATS0dv': 0.0,
  'ATS8dv': -0.0,
  'ATS0s': 0.0,
  'ATS1s': -0.0,
  'ATS0Z': 0.0,
  'ATS1Z': 0.0,
  'ATS3Z': -0.0,
  'ATS4Z': 0.0,
  'ATS5Z': 0.0,
  'ATS6Z': -0.0,
  'ATS8Z': -0.0,
  'ATS0p': 0.0,
  'AATS0dv': 0.0,
  'AATS6dv': 0.0,
  'AATS0d': 0.0,
  'AATS4d': 0.0,
  'AATS5d': 0.0,
  'AATS7d': 0.0,
  'AATS8d': 0.0,
  'AATS0s': 0.0,
  'AATS1s': 0.0,
  'AATS4s': 0.0,
  'AATS5s': -0.0,
  'AATS6s': -0.0,
  'AATS7s': 0.0,
  'AATS8s': 0.0,
  'AATS0Z': 0.0,
  'AATS4Z': 0.0,
  'AATS5Z': 0.0,
  'AATS7Z': -0.0,
  'AATS8Z': 0.0,
  'AATS0v': 0.0,
  'AATS4v': 0.0,
  'AATS5v': 0.0,
  'AATS6v': 0.0,
  'AATS7v': 0.0,
  'AATS8v': 0.0,
  'AATS0se': 0.0,
  'AATS3se': 0.0,
  'AATS5se': 0.0,
  'AATS0p': 0.0,
  'AATS5p': 0.0,
  'AATS7p': 0.0,
  'AATS0i': 0.0,
  'AATS3i': -0.0,
  'AATS4i': -0.0,
  'AATS5i': 0.0,
  'AATS6i': -0.0,
  'AATS7i': 0.0,
  'AATS8i': 0.0,
  'ATSC0c': 0.0,
  'ATSC1c': -0.0,
  'ATSC2c': 0.0,
  'ATSC3c': -0.0,
  'ATSC5c': 0.0,
  'ATSC0dv': 0.0,
  'ATSC2dv': -0.0,
  'ATSC3dv': -0.0,
  'ATSC4dv': 0.0,
  'ATSC5dv': 0.0,
  'ATSC6dv': 0.0,
  'ATSC7dv': -0.0,
  'ATSC8dv': 0.0,
  'ATSC1d': -0.0,
  'ATSC3d': -0.0,
  'ATSC4d': -0.0,
  'ATSC5d': 0.0,
  'ATSC6d': 0.0,
  'ATSC7d': -0.0,
  'ATSC8d': 0.0,
  'ATSC0s': 0.0,
  'ATSC2s': -0.0,
  'ATSC3s': -0.0,
  'ATSC5s': -0.0,
  'ATSC6s': -0.0,
  'ATSC7s': 0.0,
  'ATSC1Z': 0.0,
  'ATSC2Z': -0.0,
  'ATSC4Z': -0.0,
  'ATSC6Z': 0.0,
  'ATSC7Z': -0.0,
  'ATSC8Z': -0.0,
  'ATSC1v': -0.0,
  'ATSC2v': 0.0,
  'ATSC3v': -0.0,
  'ATSC5v': -0.0,
  'ATSC6v': 0.0,
  'ATSC7v': 0.0,
  'ATSC1se': -0.0,
  'ATSC2se': -0.0,
  'ATSC6se': -0.0,
  'ATSC8se': -0.0,
  'ATSC0pe': 0.0,
  'ATSC2pe': -0.0,
  'ATSC3pe': -0.0,
  'ATSC5pe': 0.0,
  'ATSC6pe': 0.0,
  'ATSC7pe': 0.0,
  'ATSC1p': -0.0,
  'ATSC2p': 0.0,
  'ATSC3p': -0.0,
  'ATSC4p': -0.0,
  'ATSC5p': -0.0,
  'ATSC6p': 0.0,
  'ATSC7p': -0.0,
  'ATSC8p': 0.0,
  'ATSC0i': 0.0,
  'ATSC2i': -0.0,
  'ATSC3i': -0.0,
  'ATSC5i': 0.0,
  'ATSC6i': -0.0,
  'ATSC8i': 0.0,
  'AATSC0dv': 0.0,
  'AATSC1dv': -0.0,
  'AATSC3dv': -0.0,
  'AATSC4dv': -0.0,
  'AATSC5dv': 0.0,
  'AATSC6dv': 0.0,
  'AATSC7dv': -0.0,
  'AATSC8dv': 0.0,
  'AATSC8d': 0.0,
  'AATSC0s': 0.0,
  'AATSC8s': -0.0,
  'AATSC1Z': 0.0,
  'AATSC4Z': 0.0,
  'AATSC5Z': -0.0,
  'AATSC6Z': -0.0,
  'AATSC7Z': -0.0,
  'AATSC8Z': 0.0,
  'AATSC0v': 0.0,
  'AATSC3v': -0.0,
  'AATSC4v': -0.0,
  'AATSC5v': 0.0,
  'AATSC6v': 0.0,
  'AATSC7v': 0.0,
  'AATSC8v': 0.0,
  'AATSC0i': 0.0,
  'AATSC3i': -0.0,
  'AATSC4i': 0.0,
  'AATSC5i': -0.0,
  'AATSC6i': 0.0,
  'AATSC7i': -0.0,
  'AATSC8i': 0.0,
  'MATS2c': -0.0,
  'MATS3c': -0.0,
  'MATS1s': -0.0,
  'MATS6s': -0.0,
  'MATS6Z': 0.0,
  'MATS8se': 0.0,
  'MATS7pe': -0.0,
  'MATS8pe': 0.0,
  'GATS1c': 0.0,
  'GATS2c': -0.0,
  'GATS3c': -0.0,
  'GATS4c': -0.0,
  'GATS5c': -0.0,
  'GATS8c': -0.0,
  'GATS1dv': -0.0,
  'GATS2dv': -0.0,
  'GATS5dv': 0.0,
  'GATS6dv': -0.0,
  'GATS7dv': 0.0,
  'GATS8dv': -0.0,
  'GATS1d': -0.0,
  'GATS2d': -0.0,
  'GATS3d': 0.0,
  'GATS5d': 0.0,
  'GATS6d': -0.0,
  'GATS7d': -0.0,
  'GATS8d': -0.0,
  'GATS1s': -0.0,
  'GATS2s': 0.0,
  'GATS3s': -0.0,
  'GATS4s': 0.0,
  'GATS5s': 0.0,
  'GATS6s': -0.0,
  'GATS8s': 0.0,
  'GATS2Z': -0.0,
  'GATS4Z': 0.0,
  'GATS5Z': 0.0,
  'GATS6Z': -0.0,
  'GATS7Z': 0.0,
  'GATS7v': -0.0,
  'GATS8v': -0.0,
  'GATS1se': 0.0,
  'GATS2se': -0.0,
  'GATS4se': -0.0,
  'GATS5se': 0.0,
  'GATS6se': 0.0,
  'GATS7se': 0.0,
  'GATS8se': -0.0,
  'GATS1pe': -0.0,
  'GATS2pe': -0.0,
  'GATS7pe': 0.0,
  'GATS2p': -0.0,
  'GATS3p': -0.0,
  'GATS6p': 0.0,
  'GATS7p': -0.0,
  'GATS8p': 0.0,
  'GATS1i': 0.0,
  'GATS3i': -0.0,
  'GATS4i': 0.0,
  'GATS5i': 0.0,
  'GATS7i': 0.0,
  'GATS8i': 0.0,
  'BCUTd-1l': -0.0,
  'BCUTs-1h': -0.0,
  'BCUTZ-1h': 0.0,
  'BCUTse-1h': -0.0,
  'BalabanJ': -0.0,
  'SpAbs_Dzv': -0.0,
  'SpMAD_Dzv': 0.0,
  'SM1_Dzv': -0.0,
  'SpMAD_Dzse': 0.0,
  'SM1_Dzse': 0.0,
  'BertzCT': -0.0,
  'nBondsS': -0.0,
  'PNSA1': -0.0,
  'PNSA2': 0.0,
  'PNSA3': -0.0,
  'PNSA4': -0.0,
  'PNSA5': -0.0,
  'PPSA1': 0.0,
  'PPSA2': -0.0,
  'PPSA3': 0.0,
  'PPSA4': 0.0,
  'PPSA5': 0.0,
  'DPSA1': 0.0,
  'DPSA3': 0.0,
  'DPSA4': 0.0,
  'FNSA2': -0.0,
  'WNSA1': -0.0,
  'WNSA3': 0.0,
  'WNSA4': 0.0,
  'WNSA5': -0.0,
  'WPSA1': -0.0,
  'WPSA3': -0.0,
  'WPSA4': 0.0,
  'TASA': 0.0,
  'TPSA': -0.0,
  'C1SP2': -0.0,
  'C2SP2': -0.0,
  'C3SP2': 0.0,
  'C1SP3': -0.0,
  'HybRatio': -0.0,
  'Xch-5d': -0.0,
  'Xch-7d': 0.0,
  'Xch-6dv': -0.0,
  'Xch-7dv': -0.0,
  'Xc-3d': -0.0,
  'Xc-5d': 0.0,
  'Xc-3dv': -0.0,
  'Xc-5dv': 0.0,
  'Xpc-4d': 0.0,
  'NssCH2': -0.0,
  'NaaCH': -0.0,
  'NsssCH': 0.0,
  'NaasC': 0.0,
  'NaaaC': 0.0,
  'NsssN': -0.0,
  'NsOH': 0.0,
  'NaaO': -0.0,
  'NsSH': 0.0,
  'NssS': 0.0,
  'NaaS': 0.0,
  'SsssCH': -0.0,
  'SaasC': -0.0,
  'ETA_beta_s': 0.0,
  'ETA_eta': -0.0,
  'AETA_eta': -0.0,
  'AETA_eta_F': 0.0,
  'GeomShapeIndex': 0.0,
  'IC0': 0.0,
  'IC1': 0.0,
  'IC2': 0.0,
  'IC3': 0.0,
  'IC4': 0.0,
  'TIC0': 0.0,
  'TIC1': -0.0,
  'MIC0': 0.0,
  'MIC1': 0.0,
  'ZMIC1': 0.0,
  'Kier3': 0.0,
  'FilterItLogS': 0.0,
  'Mor02': -0.0,
  'Mor07': 0.0,
  'Mor11': 0.0,
  'Mor12': 0.0,
  'Mor14': -0.0,
  'Mor15': 0.0,
  'Mor16': -0.0,
  'Mor18': -0.0,
  'Mor20': 0.0,
  'Mor21': -0.0,
  'Mor22': 0.0,
  'Mor23': 0.0,
  'Mor25': -0.0,
  'Mor26': 0.0,
  'Mor30': -0.0,
  'Mor32': 0.0,
  'Mor02m': -0.0,
  'Mor03m': 0.0,
  'Mor04m': -0.0,
  'Mor05m': -0.0,
  'Mor06m': 0.0,
  'Mor07m': -0.0,
  'Mor09m': -0.0,
  'Mor10m': 0.0,
  'Mor11m': -0.0,
  'Mor15m': 0.0,
  'Mor17m': 0.0,
  'Mor19m': -0.0,
  'Mor20m': 0.0,
  'Mor22m': 0.0,
  'Mor23m': 0.0,
  'Mor24m': 0.0,
  'Mor25m': 0.0,
  'Mor28m': 0.0,
  'Mor29m': -0.0,
  'Mor31m': 0.0,
  'Mor02v': -0.0,
  'Mor03v': -0.0,
  'Mor04v': 0.0,
  'Mor06v': 0.0,
  'Mor07v': -0.0,
  'Mor08v': -0.0,
  'Mor09v': 0.0,
  'Mor10v': -0.0,
  'Mor11v': -0.0,
  'Mor12v': 0.0,
  'Mor13v': 0.0,
  'Mor14v': -0.0,
  'Mor15v': -0.0,
  'Mor16v': 0.0,
  'Mor17v': 0.0,
  'Mor18v': -0.0,
  'Mor19v': -0.0,
  'Mor20v': 0.0,
  'Mor22v': 0.0,
  'Mor24v': 0.0,
  'Mor25v': -0.0,
  'Mor26v': 0.0,
  'Mor27v': 0.0,
  'Mor28v': 0.0,
  'Mor29v': 0.0,
  'Mor30v': -0.0,
  'Mor31v': -0.0,
  'Mor32v': 0.0,
  'Mor03p': -0.0,
  'Mor08p': -0.0,
  'PEOE_VSA1': -0.0,
  'PEOE_VSA2': 0.0,
  'PEOE_VSA3': 0.0,
  'PEOE_VSA4': 0.0,
  'PEOE_VSA5': 0.0,
  'PEOE_VSA6': -0.0,
  'PEOE_VSA8': 0.0,
  'PEOE_VSA9': 0.0,
  'PEOE_VSA11': -0.0,
  'SMR_VSA4': -0.0,
  'SMR_VSA5': -0.0,
  'SMR_VSA6': 0.0,
  'SlogP_VSA2': 0.0,
  'SlogP_VSA3': -0.0,
  'SlogP_VSA4': 0.0,
  'SlogP_VSA5': 0.0,
  'SlogP_VSA8': -0.0,
  'SlogP_VSA10': 0.0,
  'EState_VSA4': -0.0,
  'EState_VSA5': 0.0,
  'EState_VSA6': -0.0,
  'EState_VSA7': 0.0,
  'EState_VSA8': -0.0,
  'VSA_EState2': -0.0,
  'VSA_EState3': 0.0,
  'VSA_EState4': 0.0,
  'VSA_EState5': 0.0,
  'VSA_EState6': -0.0,
  'VSA_EState8': -0.0,
  'PBF': -0.0,
  'bpol': -0.0,
  'n6Ring': 0.0,
  'nHRing': -0.0,
  'n5HRing': 0.0,
  'n6HRing': -0.0,
  'n6aRing': 0.0,
  'naHRing': -0.0,
  'nARing': -0.0,
  'n5ARing': -0.0,
  'SLogP': -0.0,
  'GGI1': -0.0,
  'GGI3': -0.0,
  'GGI4': -0.0,
  'GGI5': 0.0,
  'Diameter': -0.0,
  'SRW05': 0.0,
  'TSRW10': 0.0}}

feature_order ensures that the trained model sees features in the exact same order as trained.

[8]:

feature_order[:5]

[8]:

['ABC', 'ABCGG', 'nBase', 'SpDiam_A', 'nAromAtom']

[9]:

from chemml.chem import Molecule
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
import tqdm
from chemml.chem import Mordred
from sklearn.svm import SVR

mol_objs_list = []
for i, smi in enumerate(tqdm.tqdm(df["smiles"], desc="Converting SMILES to ChemML Molecule objects")):
    mol = Molecule(smi, 'smiles')
    mol.hydrogens('add')
    try:
        mol.to_xyz('MMFF', maxIters=10000, mmffVariant='MMFF94s')
        mol_objs_list.append(mol)
    except Exception as e:
        print(f"\nUnable to process SMILES: {smi}; Error: {e}")
        # self.discarded_indices.append(i)
loaded_mols = mol_objs_list
features = Mordred(ignore_3D=False).represent(loaded_mols).drop(columns=['SMILES'])[feature_order].values
targets = df["density_Kg/m3"].values


X_test = loaded_scaler.transform(features)

y_pred = loaded_model.predict(X_test)

Converting SMILES to ChemML Molecule objects: 100%|██████████| 200/200 [00:06<00:00, 29.14it/s]

Step 1: Normalizing molecule input
Step 2/3: Optimizing geometry for 3D descriptor calculation
Step 3/3: Calculating Mordred descriptors

The loaded_model object can be treated just like any model object

[10]:

loaded_model

[10]:

Lasso(alpha=np.float64(0.10000000000000002))

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Lasso

?Documentation for LassoiFitted

Parameters

	alpha alpha: float, default=1.0 Constant that multiplies the L1 term, controlling regularization strength. `alpha` must be a non-negative float i.e. in `[0, inf)`. When `alpha = 0`, the objective is equivalent to ordinary least squares, solved by the :class:`LinearRegression` object. For numerical reasons, using `alpha = 0` with the `Lasso` object is not advised. Instead, you should use the :class:`LinearRegression` object.	np.float64(0....0000000000002)
	fit_intercept fit_intercept: bool, default=True Whether to calculate the intercept for this model. If set to False, no intercept will be used in calculations (i.e. data is expected to be centered).	True
	precompute precompute: bool or array-like of shape (n_features, n_features), default=False Whether to use a precomputed Gram matrix to speed up calculations. The Gram matrix can also be passed as argument. For sparse input this option is always ``False`` to preserve sparsity.	False
	copy_X copy_X: bool, default=True If ``True``, X will be copied; else, it may be overwritten.	True
	max_iter max_iter: int, default=1000 The maximum number of iterations.	1000
	tol tol: float, default=1e-4 The tolerance for the optimization: if the updates are smaller or equal to ``tol``, the optimization code checks the dual gap for optimality and continues until it is smaller or equal to ``tol``, see Notes below.	0.0001
	warm_start warm_start: bool, default=False When set to ``True``, reuse the solution of the previous call to fit as initialization, otherwise, just erase the previous solution. See :term:`the Glossary <warm_start>`.	False
	positive positive: bool, default=False When set to ``True``, forces the coefficients to be positive.	False
	random_state random_state: int, RandomState instance, default=None The seed of the pseudo random number generator that selects a random feature to update. Used when ``selection`` == 'random'. Pass an int for reproducible output across multiple function calls. See :term:`Glossary <random_state>`.	None
	selection selection: {'cyclic', 'random'}, default='cyclic' If set to 'random', a random coefficient is updated every iteration rather than looping over features sequentially by default. This (setting to 'random') often leads to significantly faster convergence especially when tol is higher than 1e-4.	'cyclic'

Fitted attributes

Name	Type	Value
coef_ coef_: ndarray of shape (n_features,) or (n_targets, n_features) Parameter vector (w in the cost function formula).	ndarray[float64](530,)	[ 0., 0., 0.,...,-0., 0., 0.]
dual_gap_ dual_gap_: float or ndarray of shape (n_targets,) Given parameter ``alpha``, the dual gaps at the end of the optimization, same shape as each observation of y.	float64	5.519
intercept_ intercept_: float or ndarray of shape (n_targets,) Independent term in decision function.	float64	1264
n_features_in_ n_features_in_: int Number of features seen during :term:`fit`. .. versionadded:: 0.24	int	530
n_iter_ n_iter_: int or list of int Number of iterations run by the coordinate descent solver to reach the specified tolerance.	int	1000
sparse_coef_ sparse_coef_: sparse matrix of shape (n_features, 1) or (n_targets, n_features) Read-only property derived from ``coef_``.	csr_matrix[float64](1, 530)	<Compressed S...hape (1, 530)>

[11]:

targets[:5], y_pred[:5]

[11]:

(array([1328.35, 1143.8 , 1351.33, 1466.57, 1207.67]),
 array([1332.70141881, 1145.09482801, 1347.18620498, 1472.06678856,
        1204.79752648]))

[12]:

from chemml.utils import regression_metrics
regression_metrics(targets, y_pred)

[12]:

	ME	MAE	MSE	RMSE	MSLE	RMSLE	MAPE	MaxAPE	RMSPE	MPE	MaxAE	deltaMaxE	r_squared	std
0	0.455282	3.698161	29.72562	5.452121	0.000019	0.00435	0.294964	2.579643	0.433903	0.033271	33.017888	45.542634	0.996445	91.441066

[13]:

from chemml.visualization import scatter2D, SavePlot, decorator
import pandas as pd
%matplotlib inline
df_viz = pd.DataFrame()
df_viz["Actual"] = targets.reshape(-1,)
df_viz["Predicted"] = y_pred.reshape(-1,)
sc = scatter2D('r', marker='.')
fig = sc.plot(dfx=df_viz, dfy=df_viz, x="Actual", y="Predicted")

dec = decorator(title='Actual vs. Predicted',xlabel='Actual Density', ylabel='Predicted Density',
                xlim= (950,1550), ylim=(950,1550), grid=True,
                grid_color='g', grid_linestyle=':', grid_linewidth=0.5)
fig = dec.fit(fig)
# print(type(fig))
sa=SavePlot(filename='Parity',output_directory='automl_results/val/images',
            kwargs={'facecolor':'w','dpi':330,'pad_inches':0.1, 'bbox_inches':'tight'})
sa.save(obj=fig)
fig.show()

The Plot has been saved at:  .\automl_results/val/images/Parity.png

../_images/ipython_notebooks_autoML_regression_20_1.png