Importing Package

We import the package mineralML in Python.

import mineralML as mm

Importing Data

We walk through an implementation of mineralML here. Create this following file structure locally:

mineralML/
├── Chemistry.csv
└── mineralML_neuralnetwork.py

The following columns are required for this Chemistry file:

• Sample or Sample Name

• SiO₂

• TiO₂

• Al₂O₃

• FeOₜ

• MnO

• MgO

• CaO

• Na₂O

• K₂O

• Cr₂O₃

• P₂O₅

For example, here an example containing the mineral composition data in the desired input format. You can use the ChemistryTemplate.csv from the Training_Data bit of the GitHub repository to create your own. For oxides that were not analyzed or not detected, enter 0 into the cell or alternatively mineralML will fill in these empty cells with 0 values when you use the function mm.prep_df.

	Sample Name	SiO2	TiO2	Al2O3	FeOt	MnO	MgO	CaO	Na2O	K2O	P2O5	Cr2O3	Mineral
0	Z2099	42.96	1.8	14.33	4.07	0.07	17.39	12.03	3.1	0.03	nan	0.65	Amphibole
3172	SG-09-32_12	0.17	nan	nan	0.7	nan	0.13	54.32	nan	nan	40.97	nan	Apatite
4907	IgnA-2	36.9	2.31	16.4	8.2	0.08	20.6	0.03	0.71	8.79	nan	0.09	Biotite
8137	REG55-calcite-1	0	0	0	0.07	0.0457	0.0526	57.0312	0.0046	0.0216	nan	0.1757	Carbonate
9148	zk803-31-01-01	29.41	0.03	18.6	33.22	0.08	6.7	0.1	0.03	0.26	nan	nan	Chlorite
10473	17MMSG37_cpx4-1	46.8911	2.8722	6.5948	8.4685	0.1886	13.884	20.2569	0.3403	nan	nan	0.0908	Pyroxene (Cpx)
21383	CL09MB009 C2 ep 20	37.38	0.01	21.29	15.3507	0.29	0.06	23.19	0	0	nan	nan	Epidote
22647	Lw-Ec_GR1_core	38.52	0.04	21.84	27.07	1.49	5.58	5.82	0.03	0	nan	0	Garnet
27666	REG-18_kalsilite-1	37.02	0	31.43	0.35	0	0	0	0.04	29.61	nan	0.0069	Kalsilite
28666	DG-44	65.0797	nan	18.8768	0	nan	nan	0.0664	2.4138	13.194	nan	nan	Feldspar (Alkali)
34677	VS219_129 / 1 . Leu	54.7008	0.113	23.0378	0.5208	0.0064	0.0493	nan	0.0205	21.7093	nan	nan	Leucite
39803	S80_7 / 2 .	41.0155	0.0521	5.6907	5.2198	0.131	6.6045	28.6936	3.7814	0.1842	nan	nan	Melilite
40939	WS5_37	44.8	0.24	35.12	0.269	0.04	3.22	0.16	0.15	10.57	nan	0	Muscovite
42090	10_N_1	42.033	nan	32.705	0	nan	nan	0.105	15.507	7.859	nan	nan	Nepheline
43110	CN_C_Ol1	39.846	2e-05	0.01915	17.3987	0.243865	43.1267	0.21963	0.01495	0.007775	0.013685	nan	Olivine
65125	L04_N1_1	56.08	0.2769	1.868	7.21	0.1732	34.17	0.517	nan	nan	nan	0.4681	Orthopyroxene
70895	K8_plag1_rtoc	46.6657	0.0297	32.4782	0.5769	0.0013	0.2178	16.8384	1.7939	0.0031	nan	nan	Plagioclase
25650	UC1250	0.0424	49.1927	0.0757	43.9452	1.6045	2.8229	0.0139	nan	nan	nan	0.0148	Rhombohedral-Oxide
90207	E2718C-1	0.0091	98.4337	0.0236	0.2144	0.0089	nan	nan	nan	nan	nan	0.1976	Rutile
91925	OM15-6	41.2574	nan	1.34979	4.38996	nan	39.5454	nan	nan	nan	nan	0.26425	Serpentine
89207	OM08-206A_2	99.7	0	0	0.3	0.03	0	0	0.01	0.02	nan	0	SiO2-Polymorph
36205	UC1080	0.16	16.572	0.941	74.515	0.565	0.851	0.037	nan	nan	nan	0.012	Spinel-Group
96256	REG-19-titanite-1	29.3211	33.0262	1.3941	2.90473	0.0339	0.076	26.5311	0.1435	0.0327	nan	0	Titanite
104307	Tourmaline1	36.47	0.82	30.79	4.13	nan	9.52	0.74	2.36	nan	nan	nan	Tourmaline
105329	Zrn-I	32.816	0.005	0	0.007	nan	nan	0	0.008	nan	0.027	nan	Zircon
106329	20B-04	49.96	1.59	13.73	11.8	0.21	7.27	11.88	2.27	0.25	0.18	nan	Glass

For the mineral composition, mineralML asks that users provide Fe as FeO_t. To avoid ambiguity, the Chemistry file handles this by providing only one column for FeO_t.

We use the os package in Python to facilitate navigation to various files. To load the Chemistry file, you must provide the path to the CSV.

path = os.getcwd() + '/Chemistry.csv'
df_load = mm.load_df(path)
df = mm.prep_df(df_load)

mm.load_df returns df_load, an initial dataframe of all of all samples and their chemistry. mm.prep_df then prepares the loaded dataframe by filling in any nan values, ensuring all required oxide columns are present, optionally converting all Fe to FeO_t and dropping rows with fewer than n oxides.

Data Import Complete

That is all for loading your mineral chemical compositions! You are ready to get rolling with mineralML. See the example in ML Predictions for Tabular Data (mineralML_neuralnetwork.ipynb), under the big examples heading, to see how to run mineralML and export files.