Tutorials/1_Fill-Mask.md

# Use the trained astroBERT model with the fill-mask pipeline

Tutorial 1: using astroBERT with the fill-mask pipeline

# 1 - load models and tokenizer

from transformers import AutoTokenizer, BertForMaskedLM

2022-10-14 15:27:35.809315: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0

# the model path can either be the name of the Huggingface repository
remote_model_path = 'adsabs/astroBERT'
# or the local path to the directory containing model weight and tokenizer vocab
local_model_path = '../'

# make sure you load the tokenier with do_lower_case=False
astroBERT_tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path=remote_model_path,
                                                    use_auth_token=True,
                                                    add_special_tokens=False,
                                                    do_lower_case=False,
                                                   )

astroBERT_automodel_for_mlm = BertForMaskedLM.from_pretrained(pretrained_model_name_or_path=remote_model_path, 
                                                use_auth_token=True,
                                               )

Some weights of the model checkpoint at adsabs/astroBERT were not used when initializing BertForMaskedLM: ['cls.seq_relationship.bias', 'cls.seq_relationship.weight']
- This IS expected if you are initializing BertForMaskedLM from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing BertForMaskedLM from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).

# for pipeline to work you have to ensure that the model returns a dict
astroBERT_automodel_for_mlm.config.return_dict=True

from transformers import FillMaskPipeline

astroBERT_pipeline = FillMaskPipeline(model=astroBERT_automodel_for_mlm,
                                      tokenizer=astroBERT_tokenizer,
                                      task='fill-mask',
                                     )

clean_sentences = ['M67 is one of the most studied open clusters.',
'A solar twin is a star with atmospheric parameters and chemical composition very similar to our Sun.',
'The dynamical evolution of planets close to their star is affected by tidal effects',
'The Kepler satellite collected high-precision long-term and continuous light curves for more than 100,000 solar-type stars',
'The Local Group is composed of the Milky Way, the Andromeda Galaxy, and numerous smaller satellite galaxies.',
'Cepheid variables are used to determine the distances to galaxies in the local universe.',
'Jets are created and sustained by accretion of matter onto a compact massive object.',
'A single star of one solar mass will evolve into a white dwarf.',
'The Very Large Array observes the sky at radio wavelengths.',
'Elements heavier than iron are generated in supernovae explosions.',
'Spitzer was the first spacecraft to fly in an Earth-trailing orbit.',
'Galaxy mergers can occur when two (or more) galaxies collide',
'Dark matter is a hypothetical form of matter thought to account for approximately 85% of the matter in the universe.',
'The cosmic microwave background (CMB, CMBR), in Big Bang cosmology, is electromagnetic radiation which is a remnant from an early stage of the universe.',
'The Local Group of galaxies is pulled toward The Great Attractor.',
'The Moon is the only satellite of the Earth.',
'Galaxies are categorized according to their visual morphology as elliptical, spiral, or irregular.',
'Stars are made mostly of hydrogen.',
'Comet tails are created as comets approach the Sun.',
'Pluto is a dwarf planet in the Kuiper Belt.',
'The Large and Small Magellanic Clouds are irregular dwarf galaxies and are two satellite galaxies of the Milky Way.',
'The Milky Way has a supermassive black hole, Sagittarius A*, at its center.',
'Andromeda is the nearest large galaxy to the Milky Way and is roughly its equal in mass.',
'The interstellar medium is the gas and dust between stars.']

masked_sentences =  ['M67 is one of the most studied [MASK] clusters.',
'A solar twin is a star with [MASK] parameters and chemical composition very similar to our Sun.',
'The dynamical evolution of planets close to their star is affected by [MASK] effects',
'The Kepler satellite collected high-precision long-term and continuous light [MASK] for more than 100,000 solar-type stars',
'The Local Group is composed of the Milky Way, the [MASK] Galaxy, and numerous smaller satellite galaxies.',
'Cepheid variables are used to determine the [MASK] to galaxies in the local universe.',
'Jets are created and sustained by [MASK] of matter onto a compact massive object.',
'A single star of one solar mass will evolve into a [MASK] dwarf.',
'The Very Large Array observes the sky at [MASK] wavelengths.',
'Elements heavier than [MASK] are generated in supernovae explosions.',
'Spitzer was the first [MASK] to fly in an Earth-trailing orbit.',
'Galaxy [MASK] can occur when two (or more) galaxies collide',
'Dark [MASK] is a hypothetical form of matter thought to account for approximately 85% of the matter in the universe.',
'The cosmic microwave background (CMB, CMBR), in Big Bang cosmology, is electromagnetic radiation which is a remnant from an early stage of the [MASK].',
'The Local Group of galaxies is pulled toward The Great [MASK].',
'The Moon is the only [MASK] of the Earth.',
'Galaxies are categorized according to their visual morphology as [MASK], spiral, or irregular.',
'Stars are made mostly of [MASK].',
'Comet tails are created as comets approach the [MASK].',
'Pluto is a dwarf [MASK] in the Kuiper Belt.',
'The Large and Small Magellanic Clouds are irregular [MASK] galaxies and are two satellite galaxies of the Milky Way.',
'The Milky Way has a [MASK] black hole, Sagittarius A*, at its center.',
'Andromeda is the nearest large [MASK] to the Milky Way and is roughly its equal in mass.',
'The [MASK] medium is the gas and dust between stars.']

masked_words = [x for s1,s2 in zip(clean_sentences, masked_sentences) 
              for x,y in zip(s1.split(), s2.split()) if y=='[MASK]']

masked_words

['open',
 'atmospheric',
 'tidal',
 'curves',
 'Andromeda',
 'distances',
 'accretion',
 'white',
 'radio',
 'iron',
 'spacecraft',
 'mergers',
 'matter',
 'satellite',
 'planet',
 'dwarf',
 'supermassive',
 'galaxy',
 'interstellar']

results = astroBERT_pipeline(inputs=masked_sentences, 
                             top_k=3
                            )

for w, s, rs in zip(masked_words, masked_sentences,results):
    print(s)
    print('original: {}'.format(w))
    for r in rs:
        print('\t  {} {:0.2f}'.format(r['token_str'], r['score']))
    print()

M67 is one of the most studied [MASK] clusters.
original: open
      open 0.87
      globular 0.07
      star 0.03

A solar twin is a star with [MASK] parameters and chemical composition very similar to our Sun.
original: atmospheric
      fundamental 0.56
      physical 0.25
      stellar 0.05

The dynamical evolution of planets close to their star is affected by [MASK] effects
original: tidal
      tidal 0.07
      electromagnetic 0.05
      electrostatic 0.04

The Kepler satellite collected high-precision long-term and continuous light [MASK] for more than 100,000 solar-type stars
original: curves
      curves 0.43
      ##s 0.04
      conditions 0.04

The Local Group is composed of the Milky Way, the [MASK] Galaxy, and numerous smaller satellite galaxies.
original: Andromeda
      Andromeda 0.99
      M31 0.00
      Sagittarius 0.00

Cepheid variables are used to determine the [MASK] to galaxies in the local universe.
original: distances
      distances 0.79
      distance 0.21
      redshifts 0.00

Jets are created and sustained by [MASK] of matter onto a compact massive object.
original: accretion
      accretion 0.79
      infall 0.13
      fall 0.02

A single star of one solar mass will evolve into a [MASK] dwarf.
original: white
      white 0.77
      brown 0.19
      red 0.02

The Very Large Array observes the sky at [MASK] wavelengths.
original: radio
      radio 0.29
      centimeter 0.10
      all 0.09

Elements heavier than [MASK] are generated in supernovae explosions.
original: iron
      iron 0.34
      helium 0.16
      oxygen 0.07

Spitzer was the first [MASK] to fly in an Earth-trailing orbit.
original: spacecraft
      satellite 0.42
      spacecraft 0.20
      observatory 0.16

Galaxy [MASK] can occur when two (or more) galaxies collide
original: mergers
      . 0.26
      A 0.05
      1 0.04

Dark [MASK] is a hypothetical form of matter thought to account for approximately 85% of the matter in the universe.
original: matter
      energy 0.64
      Energy 0.24
      matter 0.10

The cosmic microwave background (CMB, CMBR), in Big Bang cosmology, is electromagnetic radiation which is a remnant from an early stage of the [MASK].
original: satellite
      universe 0.45
      Universe 0.26
      expansion 0.09

The Local Group of galaxies is pulled toward The Great [MASK].
original: planet
      Wall 0.96
      East 0.01
      Planet 0.00

The Moon is the only [MASK] of the Earth.
original: dwarf
      satellite 0.38
      moon 0.31
      constituent 0.07

Galaxies are categorized according to their visual morphology as [MASK], spiral, or irregular.
original: supermassive
      elliptical 0.92
      spheroidal 0.02
      irregular 0.01

Stars are made mostly of [MASK].
original: galaxy
      hydrogen 0.20
      helium 0.14
      carbon 0.12

Comet tails are created as comets approach the [MASK].
original: interstellar
      Sun 0.45
      sun 0.23
      Earth 0.19