# Typing Practice ML Model for 10fastfingers # ------------------------------------------ # This script learns which words you are most likely to mistype # and recommends practice words for you. # # How to use: # 1. Run it once (it uses simulated logs). # 2. Later replace the synthetic logs with your real typing logs. # 3. The model will adapt and generate new practice words. import random, math, csv from collections import Counter try: from sklearn.feature_extraction import DictVectorizer from sklearn.linear_model import LogisticRegression SKLEARN_AVAILABLE = True except: SKLEARN_AVAILABLE = False # -------------------------- # 1. Simulate Typing Logs # -------------------------- letters = list("abcdefghijklmnopqrstuvwxyz") hard_letters = set(['t','r','y','u','i','o']) # pretend these are harder hard_bigrams = set(['th','tr','ri','yu','oi','ht']) def simulate_log_entry(word): entries = [] elapsed = 50 for i,ch in enumerate(word): prev = word[i-1] if i>0 else '<s>' p = 0.01 if ch in hard_letters: p += 0.12 if (prev+ch) in hard_bigrams: p += 0.10 if i==0 or i==len(word)-1: p -= 0.005 is_upper = ch.isupper() if is_upper: p += 0.05 error = 1 if random.random() < p else 0 entries.append({ 'prev_char': prev.lower(), 'char': ch.lower(), 'position': i / max(1,len(word)-1), 'is_upper': int(is_upper), 'elapsed_ms': elapsed, 'error': error }) elapsed += 30 return entries sample_words = [ "the","and","you","that","practice","keyboard","accuracy","mistake", "beautiful","synchronize","probability","algorithm","statistics", "performance","improve","predict","tough","rhythm" ] logs = [] for _ in range(1000): w = random.choice(sample_words) logs.extend(simulate_log_entry(w)) # -------------------------- # 2. Feature Extraction # -------------------------- def extract_features(entry): return { 'char=' + entry['char']: 1, 'prev_char=' + entry['prev_char']: 1, 'bigram=' + entry['prev_char']+entry['char']: 1, 'pos_bucket=' + str(int(entry['position']*4)): 1, } X_dicts = [extract_features(e) for e in logs] y = [e['error'] for e in logs] # -------------------------- # 3. Train Model # -------------------------- if SKLEARN_AVAILABLE: vec = DictVectorizer(sparse=False) X = vec.fit_transform(X_dicts) clf = LogisticRegression(max_iter=1000, class_weight='balanced', solver='liblinear') clf.fit(X, y) def predict_error_prob(entry): return float(clf.predict_proba(vec.transform([extract_features(entry)]))[0,1]) print("Model: LogisticRegression") else: counts, errors = Counter(), Counter() for d,label in zip(X_dicts,y): for k in d: counts[k]+=1 if label: errors[k]+=1 def predict_error_prob(entry): feats = extract_features(entry) rates = [] for k in feats: rate = (errors[k]+1)/(counts[k]+2) if counts[k]>0 else 0.02 rates.append(rate) return sum(rates)/len(rates) print("Model: Frequency baseline") # -------------------------- # 4. Score Words # -------------------------- def word_score(word): probs = [] for i,ch in enumerate(word): prev = word[i-1] if i>0 else '<s>' entry = { 'prev_char': prev.lower(), 'char': ch.lower(), 'position': i / max(1,len(word)-1), 'is_upper': int(ch.isupper()), 'elapsed_ms': 50 } probs.append(predict_error_prob(entry)) return sum(probs) scored = [(w, word_score(w)) for w in sample_words] scored.sort(key=lambda x: -x[1]) # -------------------------- # 5. Show Results # -------------------------- print("\nTop Practice Words:\n") for w,s in scored: print(f"{w:12s} — {s:.3f}") # Save to file with open("practice_words.txt","w") as f: for w,s in scored: f.write(f"{w}\t{s:.6f}\n") print("\nSaved to practice_words.txt")