notebooks/functorial_language_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "supported-drill"
   },
   "source": [
    "# Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "id": "quick-mexican"
   },
   "outputs": [],
   "source": [
    "from discopy import Functor, Ty, Word, Id, Cup\n",
    "\n",
    "class WordBox(Word):\n",
    "    \"\"\" Word with Ty((name, cod)) as domain. \"\"\"\n",
    "    def __init__(self, name, cod):\n",
    "        super().__init__(name, cod, dom=Ty((name, cod)))\n",
    "\n",
    "\n",
    "def remove(word, sentence):\n",
    "    \"\"\"\n",
    "    Takes a WordBox and a sentence, replaces that box by an identity,\n",
    "    and every other WordBoxes by a standard Word with empty domain.\n",
    "    \"\"\"\n",
    "    def ob(x):\n",
    "        obj, = x\n",
    "        if not isinstance(obj.name, tuple):\n",
    "            return x  # Standard types get mapped to themselves.\n",
    "        if obj.name == (word.name, word.cod):\n",
    "            return word.cod\n",
    "        return Ty()  # Normal words have empty dom.\n",
    "\n",
    "    def ar(f):\n",
    "        if f == word:\n",
    "            return Id(word.cod)\n",
    "        return Word(f.name, f.cod)\n",
    "    return Functor(ob, ar)(sentence)\n",
    "\n",
    "# vocab\n",
    "s, n = Ty('s'), Ty('n')\n",
    "tv = n.r @ s @ n.l\n",
    "iv = n.r @ s\n",
    "p = s.r @ n.r.r @ n.r @ s @ n.l\n",
    "\n",
    "vocab = {\n",
    "    n: [\"dog\", \"cat\", \"mouse\", \"fox\", \"chicken\", \"grain\", \"cheese\", \"bone\", \"fish\", \"whale\", \"seal\", \"krill\", \"water\", \"land\"],\n",
    "    tv: [\"chases\", \"flees\", \"bites\", \"eats\"],\n",
    "    iv: [\"barks\", \"meows\", \"squeaks\", \"clucks\", \"chases\", \"runs\", \"swims\"],\n",
    "     p: [\"at\", \"after\", \"in\", \"on\"]\n",
    "}\n",
    "\n",
    "# grammars\n",
    "grammars = [Cup(n, n.r) @ Id(s),\n",
    "            Cup(n, n.r) @ Id(s) @ Cup(n.l, n), \n",
    "            (Id(n) @ Id(n.r) @ Cup(s, s.r) @ Id(n.r.r) @ Id(n.r) >> Id(n) @ Cup(n.r, n.r.r) @ Id(n.r) >> Cup(n, n.r)) @ Id(s) @ Cup(n.l, n)]\n",
    "\n",
    "# transitive sentences (grammars[1])\n",
    "trans_corpus = {\n",
    "    \"dog\": {\n",
    "        \"chases\": [\"cat\", \"fox\"],\n",
    "        \"bites\": [\"cat\", \"fox\", \"bone\"],\n",
    "        \"eats\": [\"bone\"]\n",
    "    },\n",
    "    \"cat\": {\n",
    "        \"chases\": [\"mouse\"],\n",
    "        \"flees\": [\"dog\"],\n",
    "        \"bites\": [\"mouse\", \"fish\"],\n",
    "        \"eats\": [\"fish\"]\n",
    "    },\n",
    "    \"mouse\": {\n",
    "        \"flees\": [\"cat\"],\n",
    "        \"bites\": [\"cheese\"],\n",
    "        \"eats\": [\"cheese\"]\n",
    "    },\n",
    "    \"fox\": {\n",
    "        \"chases\": [\"chicken\"],\n",
    "        \"flees\": [\"dog\"],\n",
    "        \"bites\": [\"chicken\"],\n",
    "        \"eats\": [\"chicken\"]\n",
    "    },\n",
    "    \"chicken\": {\n",
    "        \"flees\": [\"fox\"],\n",
    "        \"eats\": [\"grain\"]\n",
    "    },\n",
    "    \"whale\": {\n",
    "        \"eats\": [\"krill\"]\n",
    "    },\n",
    "    \"seal\": {\n",
    "        \"eats\": [\"fish\"]\n",
    "    }\n",
    "}\n",
    "\n",
    "# intransitive sentences (grammars[0, 2])\n",
    "itrans_corpus = {\n",
    "    \"dog\": {\n",
    "        \"runs\": {\n",
    "            \"on\": [\"land\"],\n",
    "            \"after\": [\"cat\", \"fox\"]\n",
    "        },\n",
    "        \"chases\": {\n",
    "            \"after\": [\"cat\", \"fox\"]\n",
    "        },\n",
    "        \"barks\": {\n",
    "            \"at\": [\"cat\", \"fox\"]\n",
    "        }\n",
    "      },\n",
    "      \"cat\": {\n",
    "          \"runs\": {\n",
    "              \"on\": [\"land\"],\n",
    "              \"after\": [\"mouse\"]\n",
    "          },\n",
    "          \"chases\": {\n",
    "              \"after\": [\"mouse\"]\n",
    "          },\n",
    "          \"meows\": { }\n",
    "      },\n",
    "      \"mouse\": {\n",
    "          \"runs\": {\n",
    "              \"on\": [\"land\"]\n",
    "          },\n",
    "          \"squeaks\": { }\n",
    "      },\n",
    "      \"fox\": {\n",
    "          \"runs\": {\n",
    "              \"on\": [\"land\"],\n",
    "              \"after\": [\"chicken\"]\n",
    "          },\n",
    "          \"chases\": {\n",
    "              \"after\": [\"chicken\"]\n",
    "          }\n",
    "      },\n",
    "      \"chicken\": {\n",
    "          \"runs\": {\n",
    "              \"on\": [\"land\"]\n",
    "          },\n",
    "          \"clucks\": { }\n",
    "      },\n",
    "      \"fish\": {\n",
    "          \"swims\": {\n",
    "              \"in\": [\"water\"]\n",
    "          }\n",
    "      },\n",
    "      \"whale\": {\n",
    "          \"swims\": {\n",
    "              \"in\": [\"water\"]\n",
    "          }\n",
    "      },\n",
    "      \"seal\": {\n",
    "          \"swims\": {\n",
    "              \"in\": [\"water\"]\n",
    "          },\n",
    "          \"runs\": {\n",
    "              \"on\": [\"land\"]\n",
    "          }\n",
    "      },\n",
    "}\n",
    "\n",
    "# data format: { sentence: (diagram, type, word : str) }\n",
    "dataset = {}\n",
    "for subj, verb_dict in trans_corpus.items():\n",
    "  subj_box = WordBox(subj, n)\n",
    "  for verb, dobjs in verb_dict.items():\n",
    "    verb_box = WordBox(verb, tv)\n",
    "    for dobj in dobjs:\n",
    "      dobj_box = WordBox(dobj, n)\n",
    "      diagram = subj_box @ verb_box @ dobj_box >> grammars[1]\n",
    "      dataset[f\"{subj} {verb} ?\"] = (remove(dobj_box, diagram), dobj_box.cod, dobj)\n",
    "      dataset[f\"{subj} ? {dobj}\"] = (remove(verb_box, diagram), verb_box.cod, verb)\n",
    "\n",
    "    \n",
    "for subj, verb_dict in itrans_corpus.items():\n",
    "  subj_box = WordBox(subj, n)\n",
    "  for verb, prep_dict in verb_dict.items():\n",
    "    verb_box = WordBox(verb, iv)\n",
    "    diagram = subj_box @ verb_box >> grammars[0]\n",
    "    dataset[f\"{subj} ?\"] = (remove(verb_box, diagram), verb_box.cod, verb)\n",
    "    dataset[f\"? {verb}\"] = (remove(subj_box, diagram), subj_box.cod, subj)\n",
    "\n",
    "    for prep, idobjs in prep_dict.items():\n",
    "      prep_box = WordBox(prep, p)\n",
    "      for idobj in idobjs:\n",
    "        idobj_box = WordBox(idobj, n)\n",
    "        diagram = subj_box @ verb_box @ prep_box @ idobj_box >> grammars[2]\n",
    "        dataset[f\"? {verb} {prep} {idobj}\"] = (remove(subj_box, diagram), subj_box.cod, subj)\n",
    "        dataset[f\"{subj} ? {prep} {idobj}\"] = (remove(verb_box, diagram), verb_box.cod, verb)\n",
    "        dataset[f\"{subj} {verb} ? {idobj}\"] = (remove(prep_box, diagram), prep_box.cod, prep)\n",
    "        dataset[f\"{subj} {verb} {prep} ?\"] = (remove(idobj_box, diagram), idobj_box.cod, idobj)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "fifty-spanking"
   },
   "source": [
    "# JAX"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "id": "generic-capitol"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:absl:No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)\n"
     ]
    }
   ],
   "source": [
    "import jax.numpy as jnp\n",
    "import jax.random as random\n",
    "from jax import grad, jit\n",
    "import jax.nn as jnn\n",
    "import jax.experimental.optimizers as jaxopt\n",
    "\n",
    "key = random.PRNGKey(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "lasting-convention"
   },
   "source": [
    "# Functorial Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "id": "metric-variance"
   },
   "outputs": [],
   "source": [
    "from functools import partial\n",
    "\n",
    "from discopy.tensor import Functor as TensorFunctor\n",
    "\n",
    "\n",
    "def random_orthogonal(key, shape):\n",
    "    \"\"\" If shape is square, uses singular value decomposition\n",
    "    to generate a random orthogonal matrix. Otherwise, returns\n",
    "    a rectangular matrix with a high rate of orthogonal rows\n",
    "    and columns. Worked well empirically. \"\"\"\n",
    "    \n",
    "    H = random.normal(key, shape)\n",
    "    u, s, vh = jnp.linalg.svd(H, full_matrices=False)\n",
    "    O = u @ vh\n",
    "    return O\n",
    "\n",
    "\n",
    "class Functorial_Model:\n",
    "    def update_func(self): raise Exception('Not implemented.')\n",
    "    def loss_func(self): raise Exception('Not implemented.')\n",
    "    def acc_func(self): raise Exception('Not implemented.')\n",
    "    def predict_func(self): raise Exception('Not implemented.')\n",
    "\n",
    "        \n",
    "class FLM(Functorial_Model):\n",
    "    \n",
    "    def __init__(self, vocab, enc_dims, initializer=random_orthogonal):\n",
    "        \n",
    "        self.vocab = vocab\n",
    "        \n",
    "        self.encoding_matrices = {\n",
    "            ty: initializer(key, shape=(len(vocab[ty]), enc_dims[ty]))\n",
    "            for ty in vocab.keys()\n",
    "        }\n",
    "\n",
    "        self.functor = self.init_functor()\n",
    "    \n",
    "    \n",
    "    # Old optimizer implementation - now using jax optimizers\n",
    "    # @partial(jit, static_argnums=(0,))\n",
    "    # def update_func(self, params, X, y, learn_rate=0.3, **kwargs):\n",
    "    #     grads = grad(self.loss_func)(params, X, y, **kwargs)\n",
    "    #     return params - learn_rate * grads\n",
    "    \n",
    "    \n",
    "    @staticmethod\n",
    "    @jit\n",
    "    def loss_func(params, X, y, l1_reg=1e-2, l2_reg=1e-3, **kwargs):\n",
    "        \"\"\" Cross Entropy Loss with Regularization\n",
    "        Uses jax's log_softmax (which implements \"logsumexp trick\")\n",
    "        for numerical stability.\n",
    "        \"\"\"\n",
    "        return (-jnp.sum(y * jnn.log_softmax(X @ params.T, -1))\n",
    "                + l1_reg*jnp.sum(params)\n",
    "                + l2_reg*jnp.linalg.norm(params, ord=\"fro\"))\n",
    "\n",
    "\n",
    "    @partial(jit, static_argnums=(0,))\n",
    "    def acc_func(self, params, X, y):\n",
    "      \"\"\"\n",
    "      Accuracy is based prediction == label.\n",
    "      Note this is not strictly correct, since there might\n",
    "      be multiple valid predictions for the same diagram,\n",
    "      while a parsing of real-world data would only yield\n",
    "      a single label.\n",
    "      \"\"\"\n",
    "      predicted_class = jnp.argmax(self.predict_func(params, X), axis=-1)\n",
    "      target_class = jnp.argmax(y, axis=-1)\n",
    "      return jnp.mean(predicted_class == target_class)\n",
    "    \n",
    "    \n",
    "    @staticmethod\n",
    "    @jit\n",
    "    def predict_func(params, X):\n",
    "        \"\"\" Softmax of Activations\n",
    "        Returns probability distribution over vocab.\n",
    "        \"\"\"\n",
    "        return jnn.softmax(X @ params.T, -1)\n",
    "    \n",
    "    \n",
    "    def init_functor(self):\n",
    "        \"\"\" Initializes the functor. \"\"\"\n",
    "        # second dimensions of encoding matrices define object mapping\n",
    "        obj_map = {\n",
    "            ty: self.encoding_matrices[ty].shape[1] for ty in self.encoding_matrices.keys()\n",
    "        }\n",
    "        obj_map.update({s: 1})\n",
    "\n",
    "        # rows of encoding matrices define word vectors\n",
    "        arr_map = { }\n",
    "        for ty in self.vocab.keys():\n",
    "            arr_map.update({\n",
    "                Word(word, ty): self.encoding_matrices[ty][self.vocab[ty].index(word)]\n",
    "                for word in self.vocab[ty]\n",
    "            })\n",
    "\n",
    "        return TensorFunctor(obj_map, arr_map)\n",
    "\n",
    "\n",
    "    def _update_functor(self, word_type):\n",
    "      \"\"\"\n",
    "      Updates the functor for a given grammatical type.\n",
    "      \"\"\"\n",
    "      self.functor.ar.update({\n",
    "                Word(word, word_type): self.encoding_matrices[word_type][self.vocab[word_type].index(word)]\n",
    "                for word in self.vocab[word_type]\n",
    "            })\n",
    "    \n",
    "\n",
    "    def update_FLM(self, word_type, new_matrix):\n",
    "      \"\"\"\n",
    "      Updates both the encoding matrix and functor for a given grammatical type.\n",
    "      Use for consistency instead of _update_functor alone.\n",
    "      \"\"\"\n",
    "      self.encoding_matrices.update({ word_type: new_matrix })\n",
    "      self._update_functor(word_type)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "worse-lighting"
   },
   "source": [
    "# Utility Functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "id": "religious-national"
   },
   "outputs": [],
   "source": [
    "def preprocess(dataset, vocab, split=0.8):\n",
    "    \"\"\"\n",
    "    Return train-test split data in the form:\n",
    "    { type: [[diagrams], [labels], [sentences]] }\n",
    "    \"\"\"\n",
    "    preproc = { t: [[], [], []] for t in vocab.keys() }\n",
    "    train = { t: [[], [], []] for t in vocab.keys() }\n",
    "    test = { t: [[], [], []] for t in vocab.keys() }\n",
    "    \n",
    "    for sentence, (diagram, word_type, word) in dataset.items():\n",
    "        preproc[word_type][0].append(diagram)\n",
    "        \n",
    "        label = jnp.array([i == vocab[word_type].index(word) for i in range(len(vocab[word_type]))])\n",
    "        preproc[word_type][1].append(label)\n",
    "\n",
    "        preproc[word_type][2].append(sentence)\n",
    "\n",
    "    # train test split\n",
    "    for word_type, [diagrams, labels, sentences] in preproc.items():\n",
    "        indices = random.permutation(key, len(diagrams))\n",
    "        max_train_idx = int(0.8*len(diagrams))\n",
    "        train_indices = indices[:max_train_idx]\n",
    "        test_indices = indices[max_train_idx:]\n",
    "        \n",
    "        train[word_type][0] = [diagrams[i] for i in train_indices]\n",
    "        train[word_type][1] = jnp.array([labels[i] for i in train_indices])\n",
    "        train[word_type][2] = [sentences[i] for i in train_indices]\n",
    "        test[word_type][0] = [diagrams[i] for i in test_indices]\n",
    "        test[word_type][1] = jnp.array([labels[i] for i in test_indices])\n",
    "        test[word_type][2] = [sentences[i] for i in test_indices]\n",
    "    \n",
    "    return train, test\n",
    "\n",
    "\n",
    "def fit(model, train_data, test_data=None, opt=jaxopt.sgd(5e-2), opt_steps=10, epochs=9, report_every=3, **kwargs):\n",
    "    \"\"\"\n",
    "    Fits the model with configurable hyperparameters.\n",
    "    Loss function also takes l1_reg and l2_reg parameters, which can be passed in via kwargs.\n",
    "    \"\"\"\n",
    "\n",
    "    for epoch in range(epochs):     \n",
    "        loss = 0\n",
    "        acc = 0\n",
    "        acc_div = 0\n",
    "        for word_type, (diagrams, labels, _) in train_data.items():\n",
    "            # from grammar to tensor\n",
    "            F = model.functor\n",
    "            X = jnp.array([F(diagram).array for diagram in diagrams])\n",
    "            # for transitive verbs\n",
    "            X = jnp.reshape(X, (len(diagrams), -1))\n",
    "            \n",
    "\n",
    "            # the update step\n",
    "            # JAX optimizers, use commented out \"update_func\" for ordinary gradient descent\n",
    "            opt_init, opt_update, get_params = opt\n",
    "            opt_state = opt_init(model.encoding_matrices[word_type])\n",
    "\n",
    "            def opt_step(step, opt_state, X, labels, **kwargs):\n",
    "              grads = grad(model.loss_func)(get_params(opt_state), X, labels, **kwargs)\n",
    "              opt_state = opt_update(step, grads, opt_state)\n",
    "              return opt_state\n",
    "\n",
    "            for step in range(opt_steps):\n",
    "              opt_state = opt_step(step, opt_state, X, labels, **kwargs)\n",
    "\n",
    "            # update the FLM\n",
    "            model.update_FLM(word_type, opt_state.packed_state[0][0])\n",
    "            \n",
    "            # --- OLD OPTIMIZATION --- #\n",
    "            # new_matrix = model.update_func(\n",
    "            #     model.encoding_matrices[word_type], X, labels, **kwargs\n",
    "            # )\n",
    "            # model.encoding_matrices.update({ word_type: new_matrix })\n",
    "            # ------------------------ #\n",
    "\n",
    "            if (epoch + 1) % report_every == 0:\n",
    "              # metrics\n",
    "              # normalize loss for train-test comparison\n",
    "              acc += model.acc_func(model.encoding_matrices[word_type], X, labels)\n",
    "              loss += model.loss_func(model.encoding_matrices[word_type], X, labels)/X.shape[0]\n",
    "              acc_div += 1\n",
    "\n",
    "        if (epoch+1) % report_every == 0:\n",
    "                print(f\"training loss after {epoch+1} epochs: {loss}\")\n",
    "                print(f\"training accuracy after {epoch+1} epochs: {acc/acc_div}\\n\")\n",
    "                if test_data:\n",
    "                  test_loss = 0\n",
    "                  test_acc = 0\n",
    "                  test_acc_div = 0\n",
    "                  for word_type, (diagrams, labels, _) in test_data.items():\n",
    "                    # same as above w/o update\n",
    "                    F = model.functor\n",
    "                    X = jnp.array([F(diagram).array for diagram in diagrams])\n",
    "                    X = jnp.reshape(X, (len(diagrams), -1))\n",
    "                    test_acc += model.acc_func(model.encoding_matrices[word_type], X, labels)\n",
    "                    test_loss += model.loss_func(model.encoding_matrices[word_type], X, labels)/X.shape[0]\n",
    "                    test_acc_div += 1\n",
    "                  print(f\"test loss: {test_loss}\")\n",
    "                print(f\"test accuracy: {test_acc/test_acc_div}\\n\")\n",
    "\n",
    "\n",
    "def show_results(model, *datasets):\n",
    "    F = model.functor\n",
    "    for dataset in datasets:\n",
    "      for word_type, (diagrams, labels, sentences) in dataset.items():\n",
    "          for diagram, label, sentence in zip(diagrams, labels, sentences):\n",
    "              x = jnp.array(F(diagram).array)\n",
    "\n",
    "              # for transitive verbs\n",
    "              x = jnp.reshape(x, (-1,))\n",
    "\n",
    "              probs = model.predict_func(\n",
    "                  model.encoding_matrices[word_type], x\n",
    "              )\n",
    "              pred = vocab[word_type][int(jnp.argmax(probs))]\n",
    "              (Word(\"?\", word_type) >> diagram).draw(figsize=(6,3))\n",
    "              actual = vocab[word_type][jnp.argmax(label)]\n",
    "              print(f\"sentence: {sentence}\\nprediction: {pred} | actual: {actual}\")\n",
    "              print(f\"\\ntop probabilities:\")\n",
    "              for i in jnp.argsort(probs)[-1:-5:-1]:\n",
    "                  print(f\"  {vocab[word_type][i]}: {probs[i]:.2f}\")\n",
    "              print(\"------------------\\n\")\n",
    "\n",
    "\n",
    "def show_results_compact(model, *datasets):\n",
    "    F = model.functor\n",
    "    for dataset in datasets:\n",
    "      for word_type, (diagrams, labels, sentences) in dataset.items():\n",
    "        for diagram, label, sentence in zip(diagrams, labels, sentences):\n",
    "            x = jnp.array(F(diagram).array)\n",
    "\n",
    "            # for transitive verbs\n",
    "            x = jnp.reshape(x, (-1,))\n",
    "\n",
    "            probs = model.predict_func(\n",
    "                model.encoding_matrices[word_type], x\n",
    "            )\n",
    "            pred = vocab[word_type][int(jnp.argmax(probs))]\n",
    "            actual = vocab[word_type][jnp.argmax(label)]\n",
    "\n",
    "            print(f\"{sentence}:  {pred} ({actual})\")\n",
    "            top_probs = ', '.join(\n",
    "                      [f\"{vocab[word_type][i]}: {probs[i]:.2f}\"\n",
    "                      for i in jnp.argsort(probs)[-1:-3:-1]]\n",
    "                      )\n",
    "            print(f\"[top probabilities: {top_probs}]\")\n",
    "            print(\"--------------\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "parallel-refund"
   },
   "source": [
    "# Trying it out!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "unknown-butler",
    "outputId": "7b8f7ecd-90ad-4f0c-8f65-cddd0cd4b20e"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training loss after 1 epochs: 3.8713934421539307\n",
      "training accuracy after 1 epochs: 0.682539701461792\n",
      "\n",
      "test loss: 5.8964338302612305\n",
      "test accuracy: 0.47196972370147705\n",
      "\n",
      "training loss after 2 epochs: 1.134953498840332\n",
      "training accuracy after 2 epochs: 0.863095223903656\n",
      "\n",
      "test loss: 2.5842647552490234\n",
      "test accuracy: 0.6030303239822388\n",
      "\n",
      "training loss after 3 epochs: 0.08608520030975342\n",
      "training accuracy after 3 epochs: 0.8948412537574768\n",
      "\n",
      "test loss: 3.1037299633026123\n",
      "test accuracy: 0.6738636493682861\n",
      "\n",
      "training loss after 4 epochs: -0.15781939029693604\n",
      "training accuracy after 4 epochs: 0.8690475821495056\n",
      "\n",
      "test loss: 0.24258899688720703\n",
      "test accuracy: 0.6284090876579285\n",
      "\n",
      "training loss after 5 epochs: -0.8782056570053101\n",
      "training accuracy after 5 epochs: 0.9007936716079712\n",
      "\n",
      "test loss: -4.0811944007873535\n",
      "test accuracy: 0.7590909004211426\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# initialize the FLM\n",
    "noun_dim = len(vocab[n]) // 2\n",
    "enc_dims = {\n",
    "    n: noun_dim,\n",
    "    tv: noun_dim ** 2,\n",
    "    iv: noun_dim,\n",
    "    p: noun_dim ** 3\n",
    "}\n",
    "# try different initializations e.g. jnn.initializers.he_normal()\n",
    "# https://jax.readthedocs.io/en/latest/jax.nn.initializers.html\n",
    "flm = FLM(vocab, enc_dims, initializer=random_orthogonal)\n",
    "\n",
    "# get training and testing data\n",
    "train_data, test_data = preprocess(dataset, vocab)\n",
    "\n",
    "# JAX optimizer\n",
    "step_size=5e-2\n",
    "# try different optimization algos e.g. jaxopt.sgd(step_size)\n",
    "# https://jax.readthedocs.io/en/latest/jax.experimental.optimizers.html\n",
    "opt = jaxopt.adam(step_size)\n",
    "\n",
    "# fit the model\n",
    "# try messing around with the hyperparameters\n",
    "fit(flm, train_data, test_data, opt=opt, opt_steps=10, l1_reg=0.1, l2_reg=5e-2, epochs=5, report_every=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "eeAQTqHr75Bi"
   },
   "source": [
    "# Testing on _Unseen_ Sentences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "apparent-crawford",
    "outputId": "a300cf06-28c1-4567-bc60-11f12b6c0003"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: ? runs after mouse\n",
      "prediction: cat | actual: cat\n",
      "\n",
      "top probabilities:\n",
      "  cat: 0.96\n",
      "  fox: 0.03\n",
      "  seal: 0.00\n",
      "  water: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: ? runs on land\n",
      "prediction: land | actual: seal\n",
      "\n",
      "top probabilities:\n",
      "  land: 0.73\n",
      "  mouse: 0.13\n",
      "  dog: 0.05\n",
      "  water: 0.04\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: dog chases after ?\n",
      "prediction: cat | actual: fox\n",
      "\n",
      "top probabilities:\n",
      "  cat: 1.00\n",
      "  fox: 0.00\n",
      "  seal: 0.00\n",
      "  land: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: ? barks at fox\n",
      "prediction: dog | actual: dog\n",
      "\n",
      "top probabilities:\n",
      "  dog: 0.88\n",
      "  land: 0.12\n",
      "  water: 0.00\n",
      "  mouse: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: whale eats ?\n",
      "prediction: cheese | actual: krill\n",
      "\n",
      "top probabilities:\n",
      "  cheese: 0.53\n",
      "  fish: 0.17\n",
      "  grain: 0.10\n",
      "  bone: 0.05\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: fox flees ?\n",
      "prediction: dog | actual: dog\n",
      "\n",
      "top probabilities:\n",
      "  dog: 0.92\n",
      "  land: 0.04\n",
      "  cheese: 0.03\n",
      "  fish: 0.01\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: dog barks at ?\n",
      "prediction: fox | actual: fox\n",
      "\n",
      "top probabilities:\n",
      "  fox: 0.95\n",
      "  cat: 0.05\n",
      "  land: 0.00\n",
      "  krill: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: seal swims in ?\n",
      "prediction: water | actual: water\n",
      "\n",
      "top probabilities:\n",
      "  water: 0.97\n",
      "  dog: 0.02\n",
      "  mouse: 0.01\n",
      "  seal: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: chicken runs on ?\n",
      "prediction: land | actual: land\n",
      "\n",
      "top probabilities:\n",
      "  land: 0.60\n",
      "  dog: 0.36\n",
      "  fox: 0.03\n",
      "  water: 0.01\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: dog bites ?\n",
      "prediction: fox | actual: bone\n",
      "\n",
      "top probabilities:\n",
      "  fox: 0.99\n",
      "  cat: 0.01\n",
      "  fish: 0.00\n",
      "  chicken: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: fox chases after ?\n",
      "prediction: chicken | actual: chicken\n",
      "\n",
      "top probabilities:\n",
      "  chicken: 1.00\n",
      "  mouse: 0.00\n",
      "  fish: 0.00\n",
      "  fox: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: cat ? fish\n",
      "prediction: eats | actual: eats\n",
      "\n",
      "top probabilities:\n",
      "  eats: 0.60\n",
      "  bites: 0.35\n",
      "  flees: 0.05\n",
      "  chases: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: dog ? cat\n",
      "prediction: bites | actual: bites\n",
      "\n",
      "top probabilities:\n",
      "  bites: 0.81\n",
      "  flees: 0.18\n",
      "  eats: 0.01\n",
      "  chases: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: chicken ? fox\n",
      "prediction: flees | actual: flees\n",
      "\n",
      "top probabilities:\n",
      "  flees: 1.00\n",
      "  bites: 0.00\n",
      "  eats: 0.00\n",
      "  chases: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: mouse ?\n",
      "prediction: clucks | actual: squeaks\n",
      "\n",
      "top probabilities:\n",
      "  clucks: 0.59\n",
      "  chases: 0.22\n",
      "  swims: 0.07\n",
      "  squeaks: 0.05\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: cat ?\n",
      "prediction: runs | actual: meows\n",
      "\n",
      "top probabilities:\n",
      "  runs: 0.30\n",
      "  meows: 0.21\n",
      "  chases: 0.19\n",
      "  squeaks: 0.11\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: fish ?\n",
      "prediction: chases | actual: swims\n",
      "\n",
      "top probabilities:\n",
      "  chases: 0.85\n",
      "  clucks: 0.05\n",
      "  runs: 0.05\n",
      "  barks: 0.03\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: fox ? after chicken\n",
      "prediction: chases | actual: chases\n",
      "\n",
      "top probabilities:\n",
      "  chases: 1.00\n",
      "  runs: 0.00\n",
      "  swims: 0.00\n",
      "  squeaks: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: fish ? in water\n",
      "prediction: swims | actual: swims\n",
      "\n",
      "top probabilities:\n",
      "  swims: 0.97\n",
      "  runs: 0.02\n",
      "  barks: 0.01\n",
      "  squeaks: 0.01\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: cat chases ? mouse\n",
      "prediction: after | actual: after\n",
      "\n",
      "top probabilities:\n",
      "  after: 0.99\n",
      "  on: 0.01\n",
      "  in: 0.00\n",
      "  at: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: mouse runs ? land\n",
      "prediction: on | actual: on\n",
      "\n",
      "top probabilities:\n",
      "  on: 1.00\n",
      "  in: 0.00\n",
      "  at: 0.00\n",
      "  after: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: cat runs ? land\n",
      "prediction: on | actual: on\n",
      "\n",
      "top probabilities:\n",
      "  on: 0.99\n",
      "  after: 0.01\n",
      "  in: 0.00\n",
      "  at: 0.00\n",
      "------------------\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sentence: dog chases ? fox\n",
      "prediction: after | actual: after\n",
      "\n",
      "top probabilities:\n",
      "  after: 1.00\n",
      "  on: 0.00\n",
      "  in: 0.00\n",
      "  at: 0.00\n",
      "------------------\n",
      "\n"
     ]
    }
   ],
   "source": [
    "show_results(flm, test_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "vaCiRxSbdEUa"
   },
   "source": [
    "# Results on Whole Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "UwNgSZ7zh_fr",
    "outputId": "95953050-2a62-44cf-ae71-51c4927c9c07"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "? runs after chicken:  fox (fox)\n",
      "[top probabilities: fox: 0.55, cat: 0.45]\n",
      "--------------\n",
      "mouse runs on ?:  dog (land)\n",
      "[top probabilities: dog: 0.69, land: 0.26]\n",
      "--------------\n",
      "seal eats ?:  fish (fish)\n",
      "[top probabilities: fish: 0.89, cheese: 0.08]\n",
      "--------------\n",
      "cat eats ?:  fish (fish)\n",
      "[top probabilities: fish: 0.77, fox: 0.13]\n",
      "--------------\n",
      "? runs after fox:  dog (dog)\n",
      "[top probabilities: dog: 0.53, water: 0.43]\n",
      "--------------\n",
      "dog runs after ?:  fox (fox)\n",
      "[top probabilities: fox: 0.94, cat: 0.06]\n",
      "--------------\n",
      "dog runs on ?:  dog (land)\n",
      "[top probabilities: dog: 0.92, land: 0.07]\n",
      "--------------\n",
      "? swims:  water (seal)\n",
      "[top probabilities: water: 0.26, dog: 0.14]\n",
      "--------------\n",
      "? squeaks:  water (mouse)\n",
      "[top probabilities: water: 0.27, dog: 0.13]\n",
      "--------------\n",
      "chicken eats ?:  cheese (grain)\n",
      "[top probabilities: cheese: 0.61, grain: 0.18]\n",
      "--------------\n",
      "fox chases ?:  cheese (chicken)\n",
      "[top probabilities: cheese: 0.62, fish: 0.19]\n",
      "--------------\n",
      "seal runs on ?:  dog (land)\n",
      "[top probabilities: dog: 0.67, land: 0.23]\n",
      "--------------\n",
      "cat chases ?:  cat (mouse)\n",
      "[top probabilities: cat: 0.39, chicken: 0.18]\n",
      "--------------\n",
      "cat runs on ?:  land (land)\n",
      "[top probabilities: land: 0.34, dog: 0.27]\n",
      "--------------\n",
      "? runs after cat:  water (dog)\n",
      "[top probabilities: water: 0.49, dog: 0.39]\n",
      "--------------\n",
      "cat runs after ?:  mouse (mouse)\n",
      "[top probabilities: mouse: 0.82, water: 0.09]\n",
      "--------------\n",
      "whale swims in ?:  water (water)\n",
      "[top probabilities: water: 0.69, dog: 0.13]\n",
      "--------------\n",
      "mouse flees ?:  cat (cat)\n",
      "[top probabilities: cat: 0.52, fish: 0.21]\n",
      "--------------\n",
      "dog eats ?:  cheese (bone)\n",
      "[top probabilities: cheese: 0.35, bone: 0.33]\n",
      "--------------\n",
      "fox runs on ?:  land (land)\n",
      "[top probabilities: land: 0.41, water: 0.29]\n",
      "--------------\n",
      "? chases after chicken:  fox (fox)\n",
      "[top probabilities: fox: 1.00, cat: 0.00]\n",
      "--------------\n",
      "? swims in water:  seal (seal)\n",
      "[top probabilities: seal: 0.86, water: 0.09]\n",
      "--------------\n",
      "fish swims in ?:  water (water)\n",
      "[top probabilities: water: 0.60, dog: 0.28]\n",
      "--------------\n",
      "cat chases after ?:  chicken (mouse)\n",
      "[top probabilities: chicken: 0.82, mouse: 0.17]\n",
      "--------------\n",
      "fox runs after ?:  mouse (chicken)\n",
      "[top probabilities: mouse: 0.76, water: 0.10]\n",
      "--------------\n",
      "fox bites ?:  chicken (chicken)\n",
      "[top probabilities: chicken: 0.52, fish: 0.23]\n",
      "--------------\n",
      "? barks at cat:  dog (dog)\n",
      "[top probabilities: dog: 0.94, land: 0.06]\n",
      "--------------\n",
      "? chases:  fox (fox)\n",
      "[top probabilities: fox: 0.69, fish: 0.12]\n",
      "--------------\n",
      "mouse bites ?:  cheese (cheese)\n",
      "[top probabilities: cheese: 0.59, fish: 0.34]\n",
      "--------------\n",
      "? chases after cat:  dog (dog)\n",
      "[top probabilities: dog: 0.98, water: 0.02]\n",
      "--------------\n",
      "? barks:  dog (dog)\n",
      "[top probabilities: dog: 0.82, land: 0.15]\n",
      "--------------\n",
      "? chases after fox:  fox (dog)\n",
      "[top probabilities: fox: 0.53, land: 0.45]\n",
      "--------------\n",
      "? chases after mouse:  fox (cat)\n",
      "[top probabilities: fox: 1.00, cat: 0.00]\n",
      "--------------\n",
      "? clucks:  chicken (chicken)\n",
      "[top probabilities: chicken: 0.96, mouse: 0.02]\n",
      "--------------\n",
      "mouse eats ?:  cheese (cheese)\n",
      "[top probabilities: cheese: 0.70, grain: 0.13]\n",
      "--------------\n",
      "cat bites ?:  fish (fish)\n",
      "[top probabilities: fish: 0.48, chicken: 0.36]\n",
      "--------------\n",
      "chicken flees ?:  fox (fox)\n",
      "[top probabilities: fox: 0.46, cat: 0.26]\n",
      "--------------\n",
      "fox eats ?:  chicken (chicken)\n",
      "[top probabilities: chicken: 0.88, grain: 0.06]\n",
      "--------------\n",
      "? meows:  water (cat)\n",
      "[top probabilities: water: 0.31, fox: 0.11]\n",
      "--------------\n",
      "dog chases ?:  fish (fox)\n",
      "[top probabilities: fish: 0.64, cheese: 0.18]\n",
      "--------------\n",
      "? runs:  seal (seal)\n",
      "[top probabilities: seal: 0.31, bone: 0.14]\n",
      "--------------\n",
      "cat flees ?:  dog (dog)\n",
      "[top probabilities: dog: 0.71, cheese: 0.21]\n",
      "--------------\n",
      "seal ? fish:  eats (eats)\n",
      "[top probabilities: eats: 1.00, bites: 0.00]\n",
      "--------------\n",
      "mouse ? cheese:  eats (eats)\n",
      "[top probabilities: eats: 0.99, bites: 0.01]\n",
      "--------------\n",
      "dog ? fox:  bites (bites)\n",
      "[top probabilities: bites: 1.00, chases: 0.00]\n",
      "--------------\n",
      "whale ? krill:  eats (eats)\n",
      "[top probabilities: eats: 0.79, flees: 0.08]\n",
      "--------------\n",
      "dog ? bone:  eats (eats)\n",
      "[top probabilities: eats: 1.00, chases: 0.00]\n",
      "--------------\n",
      "fox ? chicken:  eats (eats)\n",
      "[top probabilities: eats: 0.98, bites: 0.02]\n",
      "--------------\n",
      "chicken ? grain:  eats (eats)\n",
      "[top probabilities: eats: 0.99, bites: 0.01]\n",
      "--------------\n",
      "cat ? dog:  flees (flees)\n",
      "[top probabilities: flees: 1.00, bites: 0.00]\n",
      "--------------\n",
      "mouse ? cat:  flees (flees)\n",
      "[top probabilities: flees: 0.99, eats: 0.00]\n",
      "--------------\n",
      "cat ? mouse:  bites (bites)\n",
      "[top probabilities: bites: 0.98, flees: 0.01]\n",
      "--------------\n",
      "fox ? dog:  flees (flees)\n",
      "[top probabilities: flees: 1.00, bites: 0.00]\n",
      "--------------\n",
      "chicken ?:  clucks (clucks)\n",
      "[top probabilities: clucks: 0.98, chases: 0.02]\n",
      "--------------\n",
      "dog ?:  barks (barks)\n",
      "[top probabilities: barks: 0.98, swims: 0.01]\n",
      "--------------\n",
      "chicken ? on land:  runs (runs)\n",
      "[top probabilities: runs: 1.00, chases: 0.00]\n",
      "--------------\n",
      "whale ?:  runs (swims)\n",
      "[top probabilities: runs: 0.28, barks: 0.21]\n",
      "--------------\n",
      "mouse ? on land:  runs (runs)\n",
      "[top probabilities: runs: 1.00, chases: 0.00]\n",
      "--------------\n",
      "dog ? at cat:  barks (barks)\n",
      "[top probabilities: barks: 1.00, swims: 0.00]\n",
      "--------------\n",
      "seal ? on land:  runs (runs)\n",
      "[top probabilities: runs: 0.93, chases: 0.07]\n",
      "--------------\n",
      "cat ? on land:  runs (runs)\n",
      "[top probabilities: runs: 0.89, chases: 0.09]\n",
      "--------------\n",
      "whale ? in water:  swims (swims)\n",
      "[top probabilities: swims: 0.92, runs: 0.05]\n",
      "--------------\n",
      "dog ? at fox:  barks (barks)\n",
      "[top probabilities: barks: 1.00, swims: 0.00]\n",
      "--------------\n",
      "dog ? on land:  runs (runs)\n",
      "[top probabilities: runs: 1.00, chases: 0.00]\n",
      "--------------\n",
      "dog ? after fox:  chases (chases)\n",
      "[top probabilities: chases: 0.59, runs: 0.41]\n",
      "--------------\n",
      "fox ?:  chases (chases)\n",
      "[top probabilities: chases: 0.90, squeaks: 0.02]\n",
      "--------------\n",
      "seal ?:  runs (runs)\n",
      "[top probabilities: runs: 0.76, chases: 0.12]\n",
      "--------------\n",
      "seal ? in water:  swims (swims)\n",
      "[top probabilities: swims: 1.00, barks: 0.00]\n",
      "--------------\n",
      "cat ? after mouse:  chases (chases)\n",
      "[top probabilities: chases: 0.88, runs: 0.12]\n",
      "--------------\n",
      "fox ? on land:  runs (runs)\n",
      "[top probabilities: runs: 0.99, chases: 0.01]\n",
      "--------------\n",
      "dog ? after cat:  chases (chases)\n",
      "[top probabilities: chases: 1.00, runs: 0.00]\n",
      "--------------\n",
      "seal runs ? land:  on (on)\n",
      "[top probabilities: on: 0.99, in: 0.01]\n",
      "--------------\n",
      "seal swims ? water:  in (in)\n",
      "[top probabilities: in: 1.00, on: 0.00]\n",
      "--------------\n",
      "fish swims ? water:  in (in)\n",
      "[top probabilities: in: 1.00, after: 0.00]\n",
      "--------------\n",
      "fox runs ? land:  on (on)\n",
      "[top probabilities: on: 1.00, in: 0.00]\n",
      "--------------\n",
      "fox chases ? chicken:  after (after)\n",
      "[top probabilities: after: 1.00, on: 0.00]\n",
      "--------------\n",
      "fox runs ? chicken:  after (after)\n",
      "[top probabilities: after: 0.97, at: 0.02]\n",
      "--------------\n",
      "whale swims ? water:  in (in)\n",
      "[top probabilities: in: 0.99, after: 0.01]\n",
      "--------------\n",
      "dog runs ? cat:  after (after)\n",
      "[top probabilities: after: 0.99, at: 0.01]\n",
      "--------------\n",
      "cat runs ? mouse:  after (after)\n",
      "[top probabilities: after: 0.95, on: 0.03]\n",
      "--------------\n",
      "dog runs ? fox:  after (after)\n",
      "[top probabilities: after: 1.00, at: 0.00]\n",
      "--------------\n",
      "dog barks ? fox:  at (at)\n",
      "[top probabilities: at: 1.00, in: 0.00]\n",
      "--------------\n",
      "dog barks ? cat:  at (at)\n",
      "[top probabilities: at: 1.00, on: 0.00]\n",
      "--------------\n",
      "dog runs ? land:  on (on)\n",
      "[top probabilities: on: 1.00, in: 0.00]\n",
      "--------------\n",
      "chicken runs ? land:  on (on)\n",
      "[top probabilities: on: 1.00, in: 0.00]\n",
      "--------------\n",
      "dog chases ? cat:  after (after)\n",
      "[top probabilities: after: 1.00, on: 0.00]\n",
      "--------------\n",
      "? runs after mouse:  cat (cat)\n",
      "[top probabilities: cat: 0.96, fox: 0.03]\n",
      "--------------\n",
      "? runs on land:  land (seal)\n",
      "[top probabilities: land: 0.73, mouse: 0.13]\n",
      "--------------\n",
      "dog chases after ?:  cat (fox)\n",
      "[top probabilities: cat: 1.00, fox: 0.00]\n",
      "--------------\n",
      "? barks at fox:  dog (dog)\n",
      "[top probabilities: dog: 0.88, land: 0.12]\n",
      "--------------\n",
      "whale eats ?:  cheese (krill)\n",
      "[top probabilities: cheese: 0.53, fish: 0.17]\n",
      "--------------\n",
      "fox flees ?:  dog (dog)\n",
      "[top probabilities: dog: 0.92, land: 0.04]\n",
      "--------------\n",
      "dog barks at ?:  fox (fox)\n",
      "[top probabilities: fox: 0.95, cat: 0.05]\n",
      "--------------\n",
      "seal swims in ?:  water (water)\n",
      "[top probabilities: water: 0.97, dog: 0.02]\n",
      "--------------\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "chicken runs on ?:  land (land)\n",
      "[top probabilities: land: 0.60, dog: 0.36]\n",
      "--------------\n",
      "dog bites ?:  fox (bone)\n",
      "[top probabilities: fox: 0.99, cat: 0.01]\n",
      "--------------\n",
      "fox chases after ?:  chicken (chicken)\n",
      "[top probabilities: chicken: 1.00, mouse: 0.00]\n",
      "--------------\n",
      "cat ? fish:  eats (eats)\n",
      "[top probabilities: eats: 0.60, bites: 0.35]\n",
      "--------------\n",
      "dog ? cat:  bites (bites)\n",
      "[top probabilities: bites: 0.81, flees: 0.18]\n",
      "--------------\n",
      "chicken ? fox:  flees (flees)\n",
      "[top probabilities: flees: 1.00, bites: 0.00]\n",
      "--------------\n",
      "mouse ?:  clucks (squeaks)\n",
      "[top probabilities: clucks: 0.59, chases: 0.22]\n",
      "--------------\n",
      "cat ?:  runs (meows)\n",
      "[top probabilities: runs: 0.30, meows: 0.21]\n",
      "--------------\n",
      "fish ?:  chases (swims)\n",
      "[top probabilities: chases: 0.85, clucks: 0.05]\n",
      "--------------\n",
      "fox ? after chicken:  chases (chases)\n",
      "[top probabilities: chases: 1.00, runs: 0.00]\n",
      "--------------\n",
      "fish ? in water:  swims (swims)\n",
      "[top probabilities: swims: 0.97, runs: 0.02]\n",
      "--------------\n",
      "cat chases ? mouse:  after (after)\n",
      "[top probabilities: after: 0.99, on: 0.01]\n",
      "--------------\n",
      "mouse runs ? land:  on (on)\n",
      "[top probabilities: on: 1.00, in: 0.00]\n",
      "--------------\n",
      "cat runs ? land:  on (on)\n",
      "[top probabilities: on: 0.99, after: 0.01]\n",
      "--------------\n",
      "dog chases ? fox:  after (after)\n",
      "[top probabilities: after: 1.00, on: 0.00]\n",
      "--------------\n"
     ]
    }
   ],
   "source": [
    "show_results_compact(flm, train_data, test_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "COsjbdQvWum-"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [
    "supported-drill",
    "fifty-spanking",
    "lasting-convention",
    "worse-lighting"
   ],
   "name": "FLM_OO.ipynb",
   "provenance": [],
   "toc_visible": true
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}